Chapter 2.

Biped Model and Simulator: Configuration, Visualisation and Accuracy 15

CHAPTER 2

Biped Model and Simulator: Configuration,

Visualisation and Accuracy

2.1 Introduction

This chapter gives an introduction to the SimMechanics software, with all essential

components involved in the making and visualisation of a biped model explained. Settings,

configurations and procedures chosen for the creation of the mobile robot simulator are

mentioned here, and a simple example of getting started with SimMechanics is also provided.

The methods used to connect and control the joints of the biped model, and the types of

sensor values that are used for the biped model are discussed. The division of the complete

biped model into several subsystems for ease of understanding are also covered, and the

extremely useful capabilities of system functions that allow immediate control of the biped

model through the MATLAB command window are explored. A description of how rigid

bodies within SimMechanics can be visualised, and the effect of such visualisation on the

simulation speed are also covered. Finally, the effect of ODE solvers on the speed and

accuracy of simulation form an essential aspect of the SimMechanics environment, and hence

needs to be properly understood, through the use of a simple mathematical model.

Chapter 2. Biped Model and Simulator: Configuration, Visualisation and Accuracy 16

2.2 Simulator and biped model base configurations

2.2.1 SimMechanics overview

Developed by MathWorks, Simulink is used for modelling, simulating and analysing multi-

domain dynamic systems, with a primarily graphical block diagram interface. Within the

Simulink environment, more precisely within the Simscape component, SimMechanics

further improves the scope of systems that can be modelled by providing the tools required

for the creation of 3D mechanical systems. Instead of using mathematical operations, or

operate on signals, or programming equations, SimMechanics makes use of basic physical

principles and physical modelling blocks to represent physical components or relationships

directly. Its purpose is the engineering design and simulation of multi-body mechanical

systems, and representing the system through the use of rigid bodies, joints, constraints and

force elements, all under the rules determined by the standard Newtonian dynamics of forces

and torques. The differential algebraic equations (DAE) that characterises the system’s

behaviour are automatically created by the Simscape component, and allowed to interact

with other Simulink models, with variables for the components in the different physical

domains solved simultaneously.

3D Translational and rotational motions of mechanical systems are available for visualisation

with the suite of tools provided by SimMechanics. Rigid bodies can be defined by their mass

properties, inertia tensors, kinematic constraints, coordinate systems and with means of

initiating and measuring their motions. With such capabilities that provide gravity and

rotation calculations, SimMechanics can be used to develop active suspensions, robotics,

surgical devices, landing gears and a variety of other systems. To further enhance the

properties of the rigid bodies, the software allows for the importing of CAD assemblies,

where the geometrical and physical definitions are more easily obtained. Once within the

SimMechanics environment, the model can be combined with other Simulink functions, as
Chapter 2. Biped Model and Simulator: Configuration, Visualisation and Accuracy 17

well as other physical system modelling products within Simulink, for more complex

interactions in the multi-domain physical system. Other physical system modelling products

includes SimDriveline for vehicles modelling, SimElectronics for electrical systems,

SimHydraulics for hydraulic systems as well as foundation modelling, as shown in Figure 2.1.

Figure 2.1: Simscape environment. Simscape is an extension of the Simulink component within
MATLAB. It allows for physical system modelling by providing physical system blocks for different
engineering areas, and can communicate to other Simulink models through the automatic creation and
simultaneous solving of DAEs
Chapter 2. Biped Model and Simulator: Configuration, Visualisation and Accuracy 18

With the ease of integration with other existing toolboxes within MATLAB and Simulink

components, SimMechanics is chosen as the tool that will be used for the analysis of

movement and postures of a biped model. By making use of existing functions within the

MATLAB environment, the development of algorithms for movements and postures of a

robot model can be developed at a faster pace.

For our biped model being considered, the Bodies, Joints, and Sensors & Actuators sets will

make up the majority of the blocks used. Hence, these blocks will be looked at in more detail

in this chapter.

2.2.2 Body coordinates used for the biped simulator

Unless otherwise stated, all vectors, tensors and coordinates used throughout this report are

represented by the Cartesian coordinates system. The reference frame (RF) that we use is the

master SimMechanics inertial RF, and unless otherwise noted, all coordinates will be with

respect to this inertial RF. Sometimes, it will be called the world coordinates or the global

coordinates, and it does not necessarily have a coordinates of [0 0 0]. All the rigid bodies

within our system will be represented relative to the world coordinates, as shown in Figure

2.2, and for ease of analysis, the world coordinates will be set to [0 0 0] throughout our

research work.
Chapter 2. Biped Model and Simulator: Configuration, Visualisation and Accuracy 19

Figure 2.2: Coordinates system. The origin of a rigid body coordinates can be translated and rotated
with respect to the world coordinates

The equation representing the position of the rigid body relative to the world body at a

particular time instant is given by:

= [ −   − 
 −
 ]
(2.1)

The orientation of the rigid body can be represented using the quaternion, rotation matrix or

Euler angle representation (Goldstein, 1980). In the model considered, the rotation matrix

will be used to represent the orientation of the rigid body and is given by:

   
 
 =    
  

 



(2.2)

where, for example,   is the direction of cosine of  with respect to 

Calculations related to equations (2.1) and (2.2) are performed automatically by the

SimMechanics environment, and updated at each simulation step to reflect the motions of

the rigid bodies.

Chapter 2. Biped Model and Simulator: Configuration, Visualisation and Accuracy 20

2.2.3 Bodies and joints settings of the biped model

Even though SimMechanics models can be imported from the CAD software, several

modifications will be required for a mobile robot system. Hence, it is a good idea to

understand the different blocks involved in the model design. Bodies and joints form the

basic building blocks for any SimMechanics models, and they can take several forms, as

summarised in Figure 2.3.

Figure 2.3: Bodies and Joints in SimMechanics. A functional SimMechanics model must consist of at
least Bodies and Joints blocks
Chapter 2. Biped Model and Simulator: Configuration, Visualisation and Accuracy 21

Table 2.1 gives a more detailed description of the Bodies block that will be used in our biped

model machine environment. The machine environment can also be called the simulation

environment or simulation world.

Table 2.1: Bodies block description

Bodies Description and Configuration

Defines how the biped model will behave in the simulation environment, and
including gravity, dimensionality, analysis mode, constraint solver type, tolerances,
linearisation, and visualisation parameters. There can only be one machine
Machine
Environment environment block connected to one ground block for our biped machine. Settings
chosen for the biped machine in our environment includes:
2
Gravity vector: [0 -9.81 0] m/s

Dimensionality: 3D Only

Analysis Mode: Forward dynamics

Represents an immobile ground point at rest in the absolute world reference frame. It
Ground is considered as a body that cannot be moved under any circumstances and at least
one of these ground blocks must be present in the machine environment. It should be
noted that the ground block is not a flat surface, but rather is a reference point. As
mentioned in subsection 2.2.2, in order to have the world coordinates as [0 0 0], the
settings for this ground block should be:
Location [x, y, z]: [0 0 0] m

The body block is a basic element of a mechanical system and is defined through its
mass and inertia tensor, position and orientation in space, and multiple attached body
coordinates system that allow it to be connected to multiple joints, constraints, or
drivers. It has no DoFs but has a CoM, and is regarded as a rigid body. The body
properties will be automatically dealt with if a proper CAD assembly is imported into
Body
the SimMechanics environment, with position and orientation convention as
described in equations (2.1) and (2.2) and the inertia tensor as a real and symmetric 3-
by-3 matrix.

The body block also allows for the rigid body geometry to be defined. Based on the
properties set for the block, it can be viewed as a convex hull from body coordinates
locations, as an equivalent ellipsoid from mass properties, or from an external
graphics file using the stereolithographic (STL) format.
Chapter 2. Biped Model and Simulator: Configuration, Visualisation and Accuracy 22

While many joints are available, only a selected few is considered in our biped model. Most

of the joints are defined during the CAD design phase, and are converted to matching types

of joint in SimMechanics during the CAD assembly import phase. With respect to our biped

model, the essential types of joint to take note of are tabulated in Table 2.2.

Table 2.2: Joints block description

Joints Description and Configuration

The primary joint type used in our biped model is of
Revolute the revolute type. Each one of the 28 DoFs in the
model uses this primitive joint type to allow for
single rotational DoF about specified axis between
two bodies. The axis of rotation is specified in the
world coordinates by [x y z] and is imported during
the CAD assembly import.

This primitive joint type allows for a single

Prismatic
translational DoF along a specified axis between two
bodies. While not explicitly used for any of the DoFs
of the biped model, it is used within the brushing
and six-DoF type of joint. Hence, it helps to
understand this type of joint.

As one of the three primitive joints in

Spherical SimMechanics, the spherical joint type represents
three rotational DoFs at a single pivot point. Two
rotational DoFs specify a directional axis, whereas
the third rotational DoF specifies rotation about that
directional axis. While actuation of the spherical
block is not possible, it cannot become singular like
the gimbal joint type.

Understanding the gimbal joint type is useful in the

understanding of how to enable the biped to roam
Gimbal in the simulation world, albeit with limited freedom.
This joint is a composite of three revolute primitives
that enables three DoFs. A joint with three revolute
primitives becomes singular if two or three of the
rotation axes becomes parallel to create a “gimbal
lock”. In such cases, the SimMechanics simulation
will not be allowed to continue.
Weld No DoFs are available with a weld type of joint. It
locks two bodies rigidly to each other, with no
possible relative motion. The weld is used to create
a horizontal floor surface that is welded to a
reference point, more precisely, the ground block.
Chapter 2. Biped Model and Simulator: Configuration, Visualisation and Accuracy 23

A composite joint with three translational DoFs as three prismatic primitives and
three rotational DoFs as three revolute primitives, it allows for the biped model to
move within the environment with limited freedom. In order to prevent the joints
from becoming singular, none of the prismatic primitives or revolute primitives
should be parallel and create a “gimbal lock”, as simulation errors will then occur.
Brushing The “gimbal lock” can occur when the biped model is required to rotate more than
90° in any direction from its default position. In other words, while the biped model
will be able to move forward and sideways within the simulation environment, it
must not turn backwards. This joint is useful if there is a need to fix part of the biped
body such that it does not move. It allows for all six DoFs to be actuated, unlike the
six-DoF joint type. With world coordinates reference, the initial axis of actions [x y z]
settings for brushing joint were set to:

Prismatic X: [1 0 0], Prismatic Y: [0 1 0], Prismatic Z: [0 0 1]

Revolute X: [1 0 0], Revolute Y: [0 1 0], Revolute Z: [0 0 1]

Representing a composite joint with three translation DoFs as three prismatic

primitives and three rotational DoFs as one spherical primitive, the six-DoF joint type
is similar to the brushing joint type, except that the rotational DoFs are represented
as one spherical primitive instead of three revolute primitives. This avoids the
singularity that can occur in the brushing joint type when two or three of the
Six-DoF
revolute joints become parallel. Singularity still occurs if two or three of the
translational axes become parallel. Depending on the needs of the simulation, such
as the movement expected by the biped model, either the six-DoF joint type or the
brushing joint type will be chosen to allow for biped movement in the simulation
world. With world coordinates reference, the initial axis of actions [x y z] settings for
brushing joint were set to:

Prismatic X: [1 0 0], Prismatic Y: [0 1 0], Prismatic Z: [0 0 1]

Spherical cannot be actuated or given axes settings

Chapter 2. Biped Model and Simulator: Configuration, Visualisation and Accuracy 24

2.2.4 Sensors and actuators configurations of the biped model

Unlike CAD software that does not provide the required tools for the designing of controllers,

SimMechanics provides tools like sensors and actuators, shown in Figure 2.4, to allow

controlled and closed-looped motions.

Figure 2.4: Sensors & actuators. Joints and Bodies can be controlled using actuators, and sensors
provide feedback related to the specific joints or bodies

The most commonly used blocks from the sensors and actuators set in our biped model are

explained in Table 2.3.

Chapter 2. Biped Model and Simulator: Configuration, Visualisation and Accuracy 25

Table 2.3: Sensors & Actuators block description

Sensors &
Description and Configuration
Actuators
Body Actuator Given a rigid body, it is possible to apply a force for translational motion or
torque for rotational motion on a specific point of the rigid body. This block will
be used within the biped model in cases where a force needs to be applied to
the biped model, for example, a push on the back of the model to see how it
behaves and how it falls down.

If there is a need to track a specific point on a rigid body, the body sensor is
Body Sensor ideal for the task. With respect to the world coordinates system, it provides
information like position, velocity, angular velocity, acceleration, angular
acceleration and the rotation matrix for the orientation.

The use of position tracking and velocity tracking for our model is essential, as
the ground surface and ground friction will be modelled based on these two
measurements.

A joint between two bodies represent the relative DoF between them, and in
order to control their relative motion, a joint actuator block is required. There
are two methods of actuation with the use of this block and they will affect the
way the control of the actuation is modelled in our biped model. The method of
actuation can be done through a generalised force or through a motion.

Joint Actuator A generalised force: The actuation signal is interpreted as a force between the
two bodies if they are connected by a prismatic primitive. If the joint between
the two bodies is of the revolute type, the signal to the joint is interpreted as a
torque signal.

A motion: With the connection being of the prismatic type, the position,
velocity and acceleration for the linear position motion actuation will need to be
specified. If a revolute primitive is being used, angular motion actuation will be
determined by the angular position, angular velocity and angular acceleration.

For our case, control of the joint actuator through a motion signal is preferred
for its ease of control.

Each joint sensor can measure a combination of outputs that depends on the
joint primitive that it is attached to. For translational motion through prismatic
primitive, the linear position, linear velocity and linear acceleration at a
particular simulation step can be obtained in various types of measurement
Joint Sensor units. If the joint is a revolute primitive, the angle, angular velocity and angular
acceleration can be sensed in different types of measurement units. In a
spherical primitive, measurements of the quaternion, quaternion first derivative
and quaternion second derivative are obtainable. The joint sensor also allows
for computed and reaction force to be calculated for prismatic joints and
computed and reaction torque for revolute joints. Spherical joint allows only for
reaction force and reaction torque to be obtained. Specific to the SimMechanics
environment that makes use of base body and follower body, the difference
between the reaction and computed force or torque are explained as:
Chapter 2. Biped Model and Simulator: Configuration, Visualisation and Accuracy 26

Reaction force and reaction torque are three-component vectors of the force
and torque that the joint transfers from the base body to the follower body.

Computed force is the component of the reaction force projected along the
prismatic primitive axis. It is also the force along the prismatic axis that
reproduces the follower motion with respect to the base.
Computed torque is the component of the reaction torque projected along the
revolute primitive axis. It is also the torque about the revolute axis that
reproduces the follower motion with respect to the base

For ease of understanding of the difference between reaction and computed

components, an illustration of the relationship between the reaction force and
the computed force is shown in Figure 2.5. In Figure 2.5(a), a reaction force  is
shown in two dimensions, and it is separated into two orthogonal components
called computed force  and computed force  . By applying computed force
 on the primitive axis A and computed force  on the primitive axis B, the
reaction force  can be obtained. Similarly, for Figure 2.5(b), a reaction torque
can be separated into its orthogonal components, with each component applied
on its respective primitive axis to obtain the reaction torque  . For simplicity,
only one axis is shown.

With regards to our model, particularly for the revolute joints representing the
28 DoFs, the useful set of sensor data would include the angle, angular velocity
and computed torque. The reaction torque is not required as only the torque
along the revolute axis is essential for us outside of the simulation world.

The computed torque provided will also have a direction sign, depending on
which is the follower body and which is the base body. The sign is of no
importance to us, with the focus placed on the magnitude instead.

(a) (b)

Figure 2.5: Joint reaction and computed values (a) Reaction and computed force on a prismatic joint
(b) Reaction and computed torque on a revolute joint
Chapter 2. Biped Model and Simulator: Configuration, Visualisation and Accuracy 27

2.2.5 Simulator settings and biped analysis mode

As mentioned in subsection 2.2.3, the machine environment block determines how the

machine should behave in the simulation environment. Given the importance of the

simulation environment and machine behaviour, some details about the settings handled by

this machine block will now be explored.

Starting with a simple case of machine dimensionality, there are two main choices: “2D only”

and “3D only”. For a 3D model, like our biped that can move in 3D, the obvious choice is the

“3D only”. Choosing “2D only” allows for movement in 2D and any attempt to make the

model move in the third dimension will cause the whole simulation to stop with an error.

Given SimMechanics uses ODE solvers to solve Newton’s equations by integrating applied

forces and torques, and obtaining the resulting motions, the possibility exists for position

errors to occur between bodies connected by prismatic or revolute joints. These position

errors increase as the accuracy of the ODE solvers is reduced. By setting linear assembly

tolerance for prismatic joints and angular assembly tolerance for the revolute joints, the

simulation can continue provided that the position error is within the defined tolerance

value. We set linear assembly tolerance to 0.025e-3 metres and angular assembly tolerance to

1e-3 radians to allow the simulation to continue without significant deformation of the model

due to joint position errors.

Next, the analysis mode (Moler, 2004) needs to be chosen, and it defines the type of solution

for the machine’s motion. The four options presented for this parameter are described in

Table 2.4, and given that our focus is not on the analysis mode, detailed descriptions are

avoided.
Chapter 2. Biped Model and Simulator: Configuration, Visualisation and Accuracy 28

Table 2.4: Analysis mode

Analysis Mode Description and Comments

Given forces, torques and initial conditions that are applied to motion
driving actuators, the resulting positions and velocities of the rigid bodies
within the system are obtained. This is explained through Newton’s law of
motion, whereby the accelerations of the bodies’ positions are directly
related to the forces and torques applied to the bodies. In addition to having
Forwards Dynamics valid values for masses and inertia tensors, the set of initial conditions,
namely initial positions and velocities, are required in order to produce
complete solutions for motions. With forward dynamics chosen as the
method of analysis for our biped model, there will be a need to understand
the relationship between acceleration and torque when the joint actuators
for the biped model are modelled. This will be covered in section 4.3.

This method of analysis is used in cases where it is required to determine the

forces and torques needed by the actuators in order to perform user-defined
motions. This option can only be used for open topology systems, and our
biped model does not satisfy this topology requirement. The topology of a
Inverse Dynamics
machine model is defined as the global connectivity of all its bodies through
joints, constraints or drivers. An open topology is an open machine, meaning
that it has no closed loops. An open topology and a closed topology are
illustrated in Figure 2.6 as examples.

Similar to inverse dynamics, where the forces and torques are determined in
order to make the actuators produce user-specified motions, with the
exception that it is used for closed topologies instead of open topologies.
Even though our biped model is a closed loop machine, this mode of analysis
Kinematics is not used as one of the independent DoFs is not kinematically actuated,
and hence making the mechanical model action not completely specified.
The independent DoF in question is the joint that links the biped model to
the ground block, and which is the one responsible for allowing the biped to
move within the simulation world.

A variant of forward dynamics, trimming is used in order to find solutions for

inputs, outputs, states, and state derivatives that satisfy conditions set by
the user. Examples include finding steady-state solutions where some or all
of the derivatives of a system’s states are zero, and it is generally used to
Trimming
analyse steady state motions. Once steady or equilibrium states in
mechanical motions are found, they are used as the starting point for
linearisation analysis. This analysis mode is not usable with our biped model,
with user specified motions not properly carried out if used.
Chapter 2. Biped Model and Simulator: Configuration, Visualisation and Accuracy 29

After consideration of the four modes of analysis for the model, the only option for the biped

model is the forwards dynamics mode, and which is also the most flexible mode among the

four and can be used with any topologies. Forward dynamics can be used to analyse inverse

dynamics and kinematics when required, though it is less efficient than inverse dynamics and

kinematics and takes more time when computing the forces on a system.

(a)

(b)

Figure 2.6: Open and closed topologies. The topology of a machine diagram is the global connectivity
of all its elements, with bodies as elements and its connections being joints, constraints and drivers (a)
A simple pendulum modelled as an open topology (b) A four bar mechanism modelled as a closed
topology
Chapter 2. Biped Model and Simulator: Configuration, Visualisation and Accuracy 30

2.2.6 Physics engine verification in SimMechanics

SimMechanics is a simulation tool that does not contain a complete set of physics engine,

with only the presence of a gravity vector to determine the motion of rigid body machines.

The physics engine needs to be taken care of by the user through the use of other Simulink

blocks. It is very important to understand how to model some components for physics

effects, as without such knowledge, the creation of a mobile biped system within

SimMechanics is impossible. In Figure 2.7, an example is shown on how to model air drag on

a projectile motion, and our horizontal ground surface and ground friction modelling method

comes from the idea provided by this air drag modelling.

Figure 2.7: Modelling air drag. In SimMechanics, several physics effects have to be modelled as they
are not readily available
Chapter 2. Biped Model and Simulator: Configuration, Visualisation and Accuracy 31

The default in SimMechanics is the presence of a gravity vector, defined as a customisable

value of [0 -9.81 0] m/s2 in the x-y-z Cartesian system. With reference to Figure 2.7, having a

value of zero for the “Drag Coeff” block is similar to having the default simulation

environment of SimMechanics. In such case, suppose an initial velocity of 100 m/s in the

horizontal x-axis and 75 m/s in the vertical y-axis is given to a point particle with a mass of

100 kg. SimMechanics will make use of ODE solvers to find the motion of the particle. The

particle itself is considered to be two prismatic primitives in the x-axis and y-axis for 2D

motion. The case where the input to the “Drag Coeff” block is set to 0.5 is also simulated for

the same initial velocities and the combined results are as shown in Figure 2.8.

Projectile motion
350
X: 759.6 Air drag coefficient = 0.0
Y: 286.7
300 Air drag coefficient = 0.5

250
y displacement (m)

200

150

100

50
X: 1529
Y: 8.514e-012
0
0 200 400 600 800 1000 1200 1400 1600
x displacement (m)

Figure 2.8: Projectile motion. With an air drag coefficient of 0.0, the simulation is same as if
SimMechanics is running only with its default physics effect that includes only gravity. With an air drag
coefficient of 0.5, the motion is dependent on how the user has modelled the physics effect

In order to check whether the gravity vector is working properly in the SimMechanics

environment while the ODE solvers are being used, a manual check involving basic projectile

equations (Hayen, 2003) is considered in finding the maximum vertical displacement, ℎ .

The maximum vertical displacement ℎ for a point particle in space can be obtained by

solving the energy conservation equation

Chapter 2. Biped Model and Simulator: Configuration, Visualisation and Accuracy 32

 = 
(2.3)
1
! = "ℎ
2
(2.4)

!
ℎ =
(2.5)
2"

where  and  are the kinetic energy and potential energy of the point particle

respectively. Mass  of the point particle is 100 kg, with the y-component of the initial

velocity being  = 75 m/s and the gravity value " being 9.81 m/s2.

The result is ℎ = 286.6972477 m, which is what the ODE solver within SimMechanics has

been able to match closely. It should be noted here that the accuracy of the results and

smoothness of the plot in SimMechanics depends on the ODE solver and simulation step size

used, and this is explained in further detail in section 2.5.

It is also noted from Figure 2.7 that there exists a “PositionSensor” subsystem that tells the

simulation to stop when the particle falls below the horizontal ground reference value of 0

during a particular simulation time step. This further reinforces the conclusion that

SimMechanics does not have a readily available physics engine to model a horizontal ground

surface, as the particle would have continued its downward fall until the simulation was

manually stop, if such “PositionSensor” subsystem was not included.

From this example of projectile motion and air drag modelling, it was concluded that several

physics effects, particularly ground surface friction and collision detection, would have to be

modelled with the use of Simulink blocks in order to have a functional mobile robot

simulator. As for the gravity and motion aspect, SimMechanics is able to deal with it in an

accurate manner, while providing us with necessary visualisation data.

Chapter 2. Biped Model and Simulator: Configuration, Visualisation and Accuracy 33

2.2.7 Dividing the biped model into subsystems

In the process of making the biped model in SimMechanics, there will be many instances

when several blocks will be grouped together to form a subsystem. An example could be the

grouping of the blocks for the right wrist, right elbow and right shoulder into a subsystem

called the right upper limb. For our biped model, Figure 2.9 shows how the different biped

body parts are manually divided into six subsystems. In this way, it is easier to understand

which actuators, sensors or body parts belong to which section, as well as making it easier to

remove limbs from the model for design updates or testing purposes.

Figure 2.9: Division of the biped model into subsystems. It becomes easier to understand and manage
the biped model if the body parts are divided several distinct groups
Chapter 2. Biped Model and Simulator: Configuration, Visualisation and Accuracy 34

The overall process in creating a subsystem is illustrated in Figure 2.10, as well as the useful

case of masking the custom subsystems that have been created. Several of the subsystems

within our biped model are masked to make these custom-made blocks easily accessible

through the Simulink library. Modifying the masked custom-made blocks through the library

causes the update to be propagated throughout the model, thereby eliminating the error-

prone and tiresome process of manually updating each blocks had the custom-made

subsystem not been made available in the library.

Figure 2.10: Overall process for subsystem creation and masking. The biped model makes frequent use
of subsystem creation and masking as it facilitates understanding and management
Chapter 2. Biped Model and Simulator: Configuration, Visualisation and Accuracy 35

2.2.8 System functions for customised control of the biped model

Simulink provides many common library blocks for use during modelling. However, in many

cases, blocks that can provide the desired functions for a specific model might not be

available within the Simulink library. In the case of the biped model that is being designed,

there are no blocks that can readily meet most of our requirement. This is where system

functions (S-functions) written within S-function blocks become essential throughout the

modelling and simulation process for the biped model.

S-functions provide a powerful mechanism for extending the capabilities of the Simulink

environment, and can be written in MATLAB, C, C++ or Fortran language. It uses a special

calling syntax called the S-function application programming interface (API), and which

enables interaction with the Simulink engine. In order to get a brief understanding of S-

functions, we need to look at the mathematical relationship between the inputs, states,

outputs of a Simulink block, as these will be very helpful in creating S-functions for our biped

model.

A Simulink block consists of a set of inputs, a set of states and a set of outputs, as illustrated

in Figure 2.11(a) for continuous states and Figure 2.11(b) for discrete states.

(a) (b)

Figure 2.11: Mathematics of Simulink blocks. Outputs are a function of the simulation time, the inputs
and the states (a) Continuous system with continuous states (b) Discrete system with discrete states
Chapter 2. Biped Model and Simulator: Configuration, Visualisation and Accuracy 36

In the continuous system, Simulink solvers integrate a set of continuous states using the

following equations:

# = $% (', )* , +) (2.6)

)*- = $. (', )* , +) (2.7)

where the output # and derivative )*- are functions of the simulation time ', the current

system states )* and the input +. $% indicates the output function, while $. indicates the

derivative function.

In discrete systems, the model is determined by the following equations:

# = $% (', ). , +) (2.8)

)./0 = $1 (', ). , +) (2.9)

where the output # and the discrete system update )./0 are functions of the simulation time

', the current system states ). and the input +. $% indicates the output function, while $1

indicates the update function.

Systems can have both continuous and discrete states at the same time.
Chapter 2. Biped Model and Simulator: Configuration, Visualisation and Accuracy 37

With the exception of an S-function that keeps track of the simulation time by making use of

continuous sampling, the other S-functions within the biped model makes use of discrete

sampling, as tabulated in Table 2.5. Note that pseudonyms are used instead of the actual S-

function names in MATLAB, as these make it easier to understand the main purpose of the S-

functions in this report.

Table 2.5: S-functions state types in biped model

Pseudonym Sampling Type Brief Description

Actuator Discrete Controls the actuator parameters

Centre of Mass Discrete Calculates the CoM

Main Discrete Anticipates user commands

Sensor Discrete Monitors the sensors

Support Discrete Determines support polygon updates

Time Continuous Updates workspace with simulation time

Tracking Discrete Tracks specific parts of the biped model

With regards to our biped model, the main advantage in setting the S-function to be accessed

in a discrete manner is that it allows the rate of updates to be controlled in a predictable

manner. While the biped model will be running in a mostly continuous manner in the

simulation environment, it is not always necessary to access certain S-functions at each and

every simulation time step. Hence, by accessing certain S-functions only when required, the

simulation can be made to run at a faster speed since less processing would be need.

For example, sometimes there is a need to just examine how the biped model moves in the

simulation environment without a need for CoM or support polygon details. In such

instances, setting the next sample time hit of the S-functions dealing with the centre of mass

and support polygon to occur at a much later stage will help alleviate the processing
Chapter 2. Biped Model and Simulator: Configuration, Visualisation and Accuracy 38

requirement of the SimMechanics system. Knowing that the calculation of the CoM and

support polygon involves many computationally intensive algorithms, reducing the rate at

which they are required will speed up the simulation. Some brief numerical details on the

speed of simulation due to different sampling rates are available in Appendix A.

Before ending this explanation of the S-functions in our biped model, with the general

simulation stages illustrated in Figure 2.12, it should be noted that a Level-1 MATLAB S-

function was used, with the programming codes written in MATLAB language and a MATLAB

interpreter used during the simulations. This limited the speed of simulation, with restricted

access to rapid acceleration mode. More quantitative details on this speed aspect are

available in Appendix A. For this current research work, it is not an easy process to update the

S-functions to make use of the other possible S-function implementations as mentioned in

Table 2.6, but it will be a useful consideration for future work, in order to have faster

simulation speed and more flexibility.

Chapter 2. Biped Model and Simulator: Configuration, Visualisation and Accuracy 39

Figure 2.12: How the Simulink engine performs simulation. Depending on the settings chosen by the
user, certain simulation steps from the general stages of S-functions are omitted during the simulation
of the biped model in the SimMechanics environment
Chapter 2. Biped Model and Simulator: Configuration, Visualisation and Accuracy 40

Table 2.6: Available S-Function implementations

Implementation Description and Comments

Level-1 MATLAB Provides a simple MATLAB interface to interact with a small portion of the S-
S-function function API

Provides access to a more extensive set of the S-function API and supports code
Level-2 MATLAB
generation. Level-2 MATLAB S-functions supersede Level-1 MATLAB S-functions,
S-function
and is the recommended implementation method in most cases

Provides the most programming flexibility. Implementation of algorithm can be as

a C MATLAB Executable (MEX) S-function or wrapper S-function to call existing C,
Handwritten C
C++, or Fortran code. Writing a new S-function requires knowledge of the S-
MEX S-function
function API and, if inlined code for the S-function needs to be generated, the
Target Language Compiler (TLC)

A graphical user interface for programming a subset of S-function functionality.

For programmers new to writing C MEX S-functions, the S-Function Builder can be
used to generate new S-functions or incorporate existing C or C++ code without
S-Function Builder
interacting with the S-function API. The S-Function Builder can also generate TLC
files for inlining the S-function during code generation with the Simulink Coder
product

A set of MATLAB commands that helps in creating an S-function to incorporate

legacy C or C++ code. Like the S-Function Builder, the Legacy Code Tool can
Legacy Code Tool generate a TLC file to inline the S-function during code generation. The Legacy
Code Tool provides access to fewer of the methods in the S-function API than the
S-Function Builder or a handwritten C MEX S-function

As a summary, S-functions allows for customised blocks within the biped model, while making

toolboxes and other functions within MATLAB more accessible. For example, each time an S-

function is called, within that S-function could include the instruction codes to access the

MATLAB toolboxes or other functions. The S-functions play an extremely important role in

customising the simulation of the biped model, without which it would not have been

possible to control the biped model during the simulation run.

Chapter 2. Biped Model and Simulator: Configuration,, Visualisation and Accuracy 41

2.3 Visualisation of rigid body dynamics

2.3.1 Overview

Animation of the body motion while the SimMechanics mo

model
del is running can be visualised in

3D. The initial default colour of a visualised rigid body is red, but the colour can be modified

independently for each one of the rigid bodies when required. There are three ways to

represent a rigid body in SimMechanic

SimMechanics: through ellipsoids, convex hulls (Michele, 2010) or

custom body geometry with external graphics file. With a Stewart platform (Stewart, 1965) as

the example, the three types of visual representation are shown in Figure 2.13
2.13. A Stewart

platform is a type of parallel robot that incorporates six prismatic actuators.

(a) (b) (c)

Figure 2.13: Three ways of representing bodies in SimMechanics. The convex hull geometry is the
initial default geometry for all visualised bodies (a) Convex hull representation (b) Ellipsoid
representation (c) Custom body geometry

2.3.2 Convex hull representation of a SimMechanics model

The convex hull geometry for a body is based on the coordinate system (CS) of the body. Each

one of these rigid bodies has at least one

on body CS at the centre of gravity (CoG), with the

possibility of adding more b

body CS for connections to joint, actuator or sensor
ensor bl
blocks. Each

body CS has an origin point, and the collection of all these CS points, in general, defines a

volume in space. The minimum outward

outward-bending
bending surface enclosing such a volume is the

convex hull of the body

ody CSs, with SimMechanics excluding the CoG CS in the creation of the

convex hull. At least four non

non-coplanar body CSs are required for a nonzero volume, with

three non-collinear body

ody CS defining a triangle and two non-coincident
coincident origins defining a line.
Chapter 2. Biped Model and Simulator: Configuration, Visualisation and Accuracy 42

One origin is displayed just as a point. To illustrate the meaning of these convex hull

creations, Figure 2.14 can be referred to.

(a)

(b)

Figure 2.14: Convex hull geometry in SimMechanics. Any rigid body with at least one CS can be
represented in convex hull geometry (a) Representation as a nonzero volume shape (b) Representation
as a triangle, line or point
Chapter 2. Biped Model and Simulator: Configuration, Visualisation and Accuracy 43

2.3.3 Ellipsoid representation of a SimMechanics model

The ellipsoid representation for a body is based on its mass properties and CoG position.

Generally, the inertia tensor of a rigid body is real and symmetric, with three real eigenvalues

(20 , 2! , 23 ) and three orthogonal eigenvectors, with the latter forming the principle axes of the

body. In the CS defined by the three axes, the inertia tensor is diagonal. When the origin of a

rigid body is at the CoG of the body, the trace of the inertia tensor is the same in such CS and

is given by

45(6) = 20 + 2! + 23 (2.10)

Every rigid body has a unique equivalent homogenous solid ellipsoid that has the same inertia

tensor. In the Cartesian system, an ellipsoid of mass  has a surface given by the equation

) ! # = !
!
8 : +; < +8 : = 1
9 9 9>
(2.11)

where 9 , 9 and 9> are the generalised radii of the ellipsoid.

Chapter 2. Biped Model and Simulator: Configuration, Visualisation and Accuracy 44

Hence, for the axis 9 ,

5[45(6) − 22 ]
9 =?
2
(2.12)

where A = 1, 2, 3

Visualisation as the equivalent ellipsoid is only possible if the principal moments (20 , 2! , 23 )

satisfiy the triangle inequalities

2! + 23 ≥ 20 (2.13)

23 + 20 ≥ 2! (2.14)

20 + 2! ≥ 23 (2.15)

Any violation of this rule leads to an imaginary generalised radius 9 . In special cases, as

tabulated in Table 2.7, the equivalent ellipsoid might not be a true ellipsoid. The labels

(A, D, E) represents the three axes (1,2,3) or (), #, =) in any order.

Chapter 2. Biped Model and Simulator: Configuration,, Visualisation and Accuracy 45

2.7: Special cases for ellipsoid geometry

Table 2.

Inertia Tensor

2A 0 0
Example

0 2D 0 
Triangle Shape in SimMechanics
Description
0 0 2E
Inequalities Visualisation Window

Ellipsoid

A true ellipsoid with nonzero

2G + 2H I 2J volume, where all axes 9 are
1 0 0 nonzero When
2H + 2J I 2G K0 1 0L
nonzero.
2J = 2G = 2H, all three 9 are
0 0 1
2J + 2G I 2H equivalent, turning the true
ellipsoid into a sphere.

Ellipse

2G + 2H = 2J
2 0 0
2H + 2J I 2G
Zero volume but nonzero
K0 1 0L area,, where one axis 9 is
0 0 1
2J + 2G I 2H
zero.

Line

2G + 2H = 2J
1 0 0
2H + 2J = 2G K0 1 0L
Zero volume and zero area,
where two axes 9 are zero.
0 0 0
2J + 2G I 2H

Point

2G + 2H = 2J
0 0 0
2H + 2J = 2G K0 0 0L
No spatial size, where all
three 9 are zero
0 0 0
2J + 2G = 2H
zero.
Chapter 2. Biped Model and Simulator: Configuration, Visualisation and Accuracy 46

2.3.4 Custom body geometry of a SimMechanics model

The custom body geometry option makes use of external graphics files of the STL format

(Stroud and Xirouchakis, 2000). STL specifies how the rigid body looks like in 3D (Rypl and

Bittnar, 2006), through the use of an arbitrary number of elementary triangles with vertices

at arbitrary points in space, and whose orientation of edges and vertices determines the

“inside” and “outside” of the surface. The usage of STL file is illustrated in Figure 2.15,

whereby a body can be represented as a closed volume mesh comprised of triangles (Zorin et

al., 1996). The quantity of triangles used for the design determines the accuracy of

representation (Szilvasi-Nagy and Matyasi, 2003) with respect to the original material, with a

higher number of triangles necessitating larger storage size and longer time to process when

rotation or translation of the body occurs.

(a) (b) (c)

Figure 2.15: Tessellation of an object. An object can be made from a tessellation of triangles, and this
tessellation information can be saved in STL files (a) An approximate representation of the object using
lesser amount of triangles (b) Better representation using an adequate amount of triangles (c) Over
tessellation can occur if more than enough triangles reasonably represent the object
Chapter 2. Biped Model and Simulator: Configuration, Visualisation and Accuracy 47

2.3.5 Effect of visualisation settings on simulation speed

To verify how simulation speed might vary when different methods of geometry

representation for visualisation are used, a Stewart platform is simulated for 100 seconds of

simulation time. The Stewart platform, as previously illustrated in Figure 2.13, is chosen due

to its complex custom body representation and relatively demanding dynamics. Before

continuing, it is worth noting that throughout this research work, there will be mention of

three different “time”. To prevent confusion, they are briefly described in Table 2.8.

Table 2.8: Definition of time

Word Description
Time, or Real Time The time taken in the real world, such as time from wrist watches

The time within the simulation world. For example, if something is simulated in
the virtual world for 100 seconds and the speed of simulation is such that it
Simulation Time
takes 2 seconds of real time to see the 1 second in the virtual world, then the
real time taken will be 200 seconds for 100 seconds of simulation time

In SimMechanics visualisation, it is possible to control the rate at which frames

are updated. The visualisation time indicates the simulation time, and not the
real time, between two frame updates for visualisation. For example, if the
Visualisation Time
visualisation time is set to 1 second, then the frame update within the
visualisation window will update after 1 second of simulation time. The real
time between frame update will vary depending on the speed of simulation.

While testing the effect of visualisation on simulation speed, the computing hardware and all

software parameters are kept constant, with the exception of the method of geometry

representation. To minimise system bottlenecks, such that the effect that can be analysed is

mainly due to the method of geometry representation in SimMechanics, a graphics

processing unit (GPU) with 1344 cores and 32 GB of system memory are used on a quad core

processor running at 3.4 GHz with 8 threads. The visualisation window size is set to 960 x

1080 pixels and the model is viewed in 3D. The real time results and central processing unit

(CPU) percentage usage rate for the 100 seconds of simulation time, with visualisation time

that includes 10 ms, 100ms, 1s and 10s, are summarised in Figure 2.16. Inherited time means
Chapter 2. Biped Model and Simulator: Configuration, Visualisation and Accuracy 48

that the visualisation time depends on the simulation step size, and no visualisation means

that the model is simulated without the need to have visualisation.

Time taken for 100 seconds of simulation time

for the Stewart platform for different visualisation time

Assuming Real Time: Simulation Time = 1:1 No Visualisation 10 s 1s 100 ms 10 ms Inherited Time

Simulation Time 100.000

No Visualisation 11.198

Custom 12.384
Ellipsoid 12.609
Visualisation Geometry

Convex hull 12.459

Custom 13.368
Ellipsoid 13.840
Convex hull 13.500

Custom 58.536
Ellipsoid 23.783
Convex hull 22.145

Custom 462.817
Ellipsoid 130.535
Convex hull 130.544

Custom 462.838
Ellipsoid 130.540
Convex hull 130.552
0 50 100 150 200 250 300 350 400 450 500
Time taken (s)

(a)
Average CPU utilisation percentage for 100 seconds of simulation time
for the Stewart platform for different visualisation time

Assuming Real Time: Simulation Time = 1:1 No Visualisation 10 s 1s 100 ms 10 ms Inherited Time

Simulation Time
No Visualisation
Custom
Ellipsoid
Visualisation Geometry

Convex hull
Custom
Ellipsoid
Convex hull
Custom
Ellipsoid
Convex hull
Custom
Ellipsoid
Convex hull
Custom
Ellipsoid
Convex hull
0 5 10 15 20 25
Average CPU utilisation (%)

(b)

Figure 2.16: Visualisation effect on simulation speed and CPU usage. Selecting the appropriate
visualisation time and method can speed up the simulations (a) Real time taken for different
visualisation times (b) Average CPU utilisation for different visualisation times
Chapter 2. Biped Model and Simulator: Configuration, Visualisation and Accuracy 49

It can be concluded that the ellipsoid representation tends to be less resource demanding but

offers no significant simulation speed gains over convex hull representations. Custom body

representation is generally more resource demanding and leads to lower simulation speed.

However, the most significant impact comes from the visualisation time. While having a

lower value for the visualisation time will generally make the simulation animation smoother,

making it too low will in fact degrade the smoothness of simulation because the bottle neck

then comes from the CPU not being able to keep up with the instructions for graphics update

while managing the ODE solvers at the same time. On a quad core processor, a CPU

utilisation of 25% indicates the application is single-threaded and fully utilising the single

core.

Hence, with respect to the visualisation of our biped model, the visualisation time will need

to be adjusted according to the desired smoothness, and there are no fixed values for it. This

is due to the visualisation time being dependent on the simulation time, which in turn can

have different ratios with the real time depending on the simulation complexity. As a

guideline, for a ratio of 1:1 for simulation time: real time, a visualisation time of 50 ms is

reasonable enough, leading to a frame per second (fps) rate of 20. Movies generally run

fluidly at 24 or 30 fps. In some cases, the visualisation will be completely disabled when not

required, in order to spend less time for the completion of the simulation. If visualisation is

still required, disabling the CoG and CS visibility option, as previously shown in Figure 2.14,

will further increase simulation speed.

Chapter 2. Biped Model and Simulator: Configuration, Visualisation and Accuracy 50

2.3.6 Simulink 3D animation as a visualisation alternative

Before ending the part on visualisation, it might be helpful to mention that recording of the

animation can also be performed, with each updated and displayed frame captured by

SimMechanics itself. Additionally, since the translational and rotational parameters for each

rigid body within a model can be obtained easily through Sensor blocks, SimMechanics

visualisation can be bypassed and a mechanical animation in a virtual world of one’s own

design can be made. The rigid bodies can be represented as virtual objects using a virtual

reality modelling language (VRML) in a world created by Simulink 3D Animation (Marrin and

Campbell, 1997; Li et al., 2011). However, our purpose is not to beautify the environment in

which the biped model is in but to get useful parameters from the biped model, and hence it

will not be covered any further but can be considered in future improvements. Figure 2.17

depicts some possibilities of combining SimMechanics, or other Simulink products, with the

Simulink 3D Animation component.

(a) (b)

(c) (d)

Figure 2.17: Simulink 3D Animation. SimMechanics visualisation can be replaced with Simulink 3D
Animation if required, with more views and details present, such as lighting effects (a) A manipulator
(b) Plane on the runway (c) Car driving simulation and animation (d) Driver’s view inside car
Chapter 2. Biped Model and Simulator: Configuration, Visualisation and Accuracy 51

2.4 Steps taken to make the biped model fully functional

2.4.1 Overview

Even though models can be imported from the CAD software, several important

modifications will be required for a mobile system. The CAD biped model that is imported

into SimMechanics does not contain any actuated joints, nor does it have any sensors

attached. Even the ground surface on which the biped will be standing needs to be created in

SimMechanics after the model import. A simple example of how to make a functional model

and simulation world is explained in this section, and the unmodified biped model that is

imported from the CAD software will go through the same processes mentioned in this

section in order to make it fully functional. The overall process that we will describe is

summarised in a simple flow chart in Figure 2.18, while Figure 2.19 will be used as the

example for the basic model to be built in SimMechanics throughout this section, and making

it into a fully functional model. By the end of this section, it should be easy to make the biped

model into a SimMechanics system that can be controlled as desired, with sensor values

obtained whenever required and not neglecting the possibility for visualisation.
Chapter 2. Biped Model and Simulator: Configuration, Visualisation and Accuracy 52

Figure 2.18: Making a fully functional model in SimMechanics. These steps will be used in explaining
how to make a fully functional model. The biped model imported from the CAD software does not
have any useful abilities and needs multiple modifications before being fully functional

Figure 2.19: Basic model to be designed using SimMechanics. The model is assumed to be a pole with
base at coordinates (0,0,0), and the base is connected to a revolute joint that is controlled through
motion. At the end of the pole at (0,2,0), a stick is connected by using another revolute joint, and
which is controlled through force
Chapter 2. Biped Model and Simulator: Configuration, Visualisation and Accuracy 53

2.4.2 Selecting and positioning blocks for the basic model

Each model that is being made using SimMechanics requires at minimum a “Machine

Environment” block that represents the machine’s mechanical settings and a “Ground” block

to represent an immobile point at rest in absolute space. Based on Figure 2.19, there are two

revolute joints and two rigid bodies present, and hence, two “Body” block representing the

rigid bodies and two “Joint” blocks for relative motions between the “Body” blocks to which

they are connected are also needed. Figure 2.20 represents the model at this stage.

Figure 2.20: Selecting and positioning the blocks. The essential blocks to create the required basic
model have been chosen and positioned in the modelling window
Chapter 2. Biped Model and Simulator: Configuration, Visualisation and Accuracy 54

2.4.3 Blocks configuration of the basic model

Most of the block settings will be left at default values and their names will not be changed

for this basic model. Table 2.9 shows the settings for the six blocks present in Figure 2.20,

followed by examples of their setting windows in Figure 2.21.

Table 2.9: Block settings of basic model

Block Settings and Comments

Gravity vector: [0 -9.81 0] m/s^2

Gravity is acting downwards in the y-axis.

Machine Dimensionality: Auto-detect

Automatic detection of whether movement in 3D is present.

Analysis mode: Forward dynamics

This analysis mode is most appropriate for the modelling of simple models, as it does
not have to abide by the rules of open or closed model topologies. Nevertheless, this
mode can be less efficient than the other modes of analysis.

Location [x,y,z]: [0 0 0] m

The world coordinates is set to [0 0 0].

Show Machine Environment port: Checked

This will allow connection to the “Machine Environment” block by enabling an

additional port in the “Ground” block.

This block, with the name set to “Revolute”, will represent the revolute joint located
at (0,0,0).

Axis of Action [x y z]: [0 0 1], Reference CS: World

The revolute joint has a DoF in the z-axis with respect to the world RF.

Number of sensor / actuator ports: 0

For now, there will not be any actuator or sensor connected to this joint. The rigid
body, with the name set to “Body”, will only be actuated by the force due to gravity,
without the presence of any frictional forces or air resistances.

This block, with the name set to “Revolute1”, will represent the revolute joint located
at (0,2,0).

Axis of Action [x y z]: [0 0 1], Reference CS: World

The revolute joint has a DoF in the z-axis with respect to the world RF.

Number of sensor / actuator ports: 0

Chapter 2. Biped Model and Simulator: Configuration, Visualisation and Accuracy 55

This block, with the name “Body”, represents the pole from Figure 2.19. The
properties of the “Body” block can be more easily obtained if SimMechanics is used
together with the CAD software. Here, the parameters are manually set to:

Mass: 1 kg
1 0 0
Inertia: eye(3) = K0 1 0L
0 0 1
Position:

It is easier to deal with world coordinates and world RF in our case, as the
coordinates given in Figure 2.19 are in terms of world coordinates

Orientation:

This block, with the name “Body1”, represents the stick from Figure 2.19. The
parameters are manually set to:

Mass: 1 kg
1 0 0
Inertia: eye(3) = K0 1 0L
0 0 1
Position:

Orientation:
Chapter 2. Biped Model and Simulator: Configuration, Visualisation and Accuracy 56

(a) (b)

(c) (d)

Figure 2.21: Block settings window. Configuration windows for the blocks involved in the basic model
(a) “Machine Environment” block (b) “Joint” block (c) “Ground” block (d) “Body” block

2.4.4 Connecting the blocks to make the basic model

The blocks from Figure 2.20 can be connected to form a system by drawing lines between

them. Some block configurations are available only after connections between different block

are made. Three important things to note for any models, as illustrated in Figure 2.22, are:

(1) Exactly one “Machine Environment” block can be connected to a “Ground” block

(2) A body can have several joints attached in order to create branches, but each joint

can only be attached between two and only two bodies.

Chapter 2. Biped Model and Simulator: Configuration, Visualisation and Accuracy 57

(3) At least a “Machine Environment” block, a “Ground” block, a “Joint” block and a

“Body” block must be present in order for the model to be valid

Figure 2.22: Valid block connections. Some rules for block connections have to be followed in order for
the simulation to be valid. For example, “Body” blocks cannot be connected directly to other “Body”
blocks, and “Joint” blocks cannot be connected directly to other “Joint” blocks. A valid connection
always include a “Joint” block in between two “Body” blocks

Finally, the complete and valid model for our basic model will be as shown in Figure 2.23.

Pressing the “Play” button will run the simulation for the specified amount of time.

Figure 2.23: Connected and valid model. The simulation can be run for the specified amount of time by
pressing the “Play” button.
Chapter 2. Biped Model and Simulator: Configuration, Visualisation and Accuracy 58

2.4.5 Visualisation of the basic model

At this stage, though the basic model can run, there is no visualisation window. Hence, the

“Simulation Configuration Parameters” need to be set to have a visual view of the model, as

shown in Figure 2.24.

Figure 2.24: Enabling the machine model visualisation. “Display machines after updating diagram” and
“Show animation during simulation” needs to be enabled for visualisation of the machine model
Chapter 2. Biped Model and Simulator: Configuration, Visualisation and Accuracy 59

With the visualisation of the basic model enabled, the machine model can be visualised, as

shown in Figure 2.25, where screenshots of the model are taken every 100 ms for the first

1.41 seconds of the simulation, in the X-Y Cartesian plane.

t = 0.01 s t = 0.11 s t = 0.21 s t = 0.31 s t = 0.41 s

t = 0.51 s t = 0.61 s t = 0.71 s t = 0.81 s t = 0.91 s

t = 1.01 s t = 1.11 s t = 1.21 s t = 1.31 s t = 1.41 s

Figure 2.25: Machine model window. The model made by the connection of SimMechanics blocks can
be visualised in the machine model window

Similar to the basic model shown in Figure 2.19, the machine model in Figure 2.25 shows the

pole and stick, as well as their CoG as defined in their respective body blocks. The animation

speed would need to be reduced as 10 seconds of simulation for such a simple model can be

accomplished in a less than half a second.

As noted, there is no controlled motion of the two revolute joints in the model at this stage,

with the motion currently determined by gravity and the mass and inertia properties of the

rigid bodies. To have more useful analysis in the field of robotics, the joints should generally

be actuated, as well as have some sensors present.

Chapter 2. Biped Model and Simulator: Configuration, Visualisation and Accuracy 60

2.4.6 Controlled motion for the basic model

The settings for the two revolute “Joint” blocks can be updated to have values as shown in

Table 2.10, whereby ports on the “Joint” blocks are created to allow for sensor blocks or an

actuator block to be connected.

Table 2.10: Enabling of sensor and actuator ports

Block Settings and Comments

Number of sensor / actuator ports: 2

A joint cannot have more than 1 “Joint Actuator” block connected to it, but it can be
connected to several” Joint Sensor” blocks.

Number of sensor / actuator ports: 2

The connections to be made between the newly added blocks are as shown in Figure 2.26.

Figure 2.26: Basic model with sensors and actuators. “Joint Actuator” blocks allow for controlled
motion of the joints while “Joint Sensor” blocks provide feedback values
Chapter 2. Biped Model and Simulator: Configuration, Visualisation and Accuracy 61

The settings for the newly added blocks can be set as indicated in Table 2.11, with other

values left at default. Such settings are generally only available for modification when the

block connections have been made.

Table 2.11: Main settings for sensors and actuators

Block Settings and Comments

Connected to block named “Revolute”

Actuate with: Motion

The joint is expecting a motion signal vector for its angle, angular velocity and angular
acceleration in terms of degrees and seconds.

Connected to block named “Revolute1”

Actuate with: Generalised Forces

The joint is expecting a torque value for actuation in terms of Newton and metres.

Connected to block named “Revolute”

Computed torque: Checked. All others unchecked

The sensor will provide the computed torque experienced by the block named
“Revolute”

Connected to block named “Revolute1”

Angular velocity: Checked. All others unchecked

The sensor will provide the angular velocity experienced by the block named
“Revolute1”

Constant value: [45 0 0]

Connected to block named “Joint Actuator”, it will indicate to the joint holding the
pole to lean 45 degrees to the left. The direction of rotation is based on the sign of
the value, and the base and follower body of the “Joint” block to which “Joint
Actuator” is connected

Constant value: -2

Connected to block named “Joint Actuator1”, it will indicate to the joint holding the
stick that a constant torque of 2 Nm is constantly being applied, with the force acting
in the clockwise direction. If the value was positive, the direction of the force would
have been in the anticlockwise direction
Chapter 2. Biped Model and Simulator: Configuration, Visualisation and Accuracy 62

2.4.7 Results from a functional basic model

Simulation of Figure 2.26 would have resulted in a scope display as shown in Figure 2.27.

Figure 2.27: Scope for simulation of basic model. The upper plot represents the computed torque, in
Nm, experienced by the joint at the base of the pole. The lower plot represents the angular velocity, in
degrees per second, experienced by the joint at the base of the stick, as the force supplied to the joint
is sometimes in phase and sometimes in anti-phase with the force due to gravity

Many of the results obtained from the simulation in SimMechanics can be quite surprising

and not straightforward to understand without simulation. In this basic model, an

increasingly faster rotation of the stick causes the torque experienced by the joint at the base

of the pole to have larger maximum amplitude, while maintaining its angle. Eventually,

SimMechanics should be able to provide very valuable analysis details for the simulation of

our biped model.

Chapter 2. Biped Model and Simulator: Configuration, Visualisation and Accuracy 63

Furthermore, after going through this whole section, it should be possible to perform more

complex modifications of the biped model that will be imported from the CAD software. An

imported model does not contain a horizontal ground surface, nor does it have any blocks for

controlled motion and sensing. A very good grasp of the basics covered here is essential in

order to make the base biped model into a fully functional one.

Finally, note that the ODE solver used was changed from the variable-step ODE solver ode45

to a fixed-step ODE solver ode8. The ODE solver is an extremely important part of

SimMechanics as it determines the speed and accuracy of the results while solving the

equation of motions for a physical model.

Chapter 2. Biped Model and Simulator: Configuration, Visualisation and Accuracy 64

2.5 Choosing suitable ODE solvers for the biped model

2.5.1 Introducing the ODE solver candidates

An ODE solver computes the dynamic system’s states at successive time steps over a

specified time span, using information provided by the Simulink model (Schwerin, 1999).

There are two broad categories of ODE solvers: fixed-step solver, which is generally required

for code generation, and variable-step solver, which is preferred in models whose states

change rapidly or which contain discontinuities (Ashino et al., 2000). Our biped model is

simulated using the variable-step solver. By using such solver, the simulation step size is

reduced when the model changes state rapidly so as to maintain accuracy, and increases

when the model state changes slowly, in order to speed up the simulation by avoiding

unnecessary steps. Within variable-step ODE solvers in SimMechanics, there are multiple

choices, as summarised in Table 2.12.

Chapter 2. Biped Model and Simulator: Configuration, Visualisation and Accuracy 65

Table 2.12: Introducing variable-step ODE solvers within Simulink

ODE One-Step Multistep

Method Used in Solver Explicit Implicit
Solver Method Method
PECE Implementation of

ode113 Adams-Bashforth-Moutlon

(Shampine, 1975)

Numerical Differentiation

ode15s Formulas (NDFs)

(Shampine, 1997)

Runge-Kutta (2,3) pair of

ode23 Bogacki & Shampine

(Bogacki and Shampine, 1989)

Second-order, modified
ode23s


Rosenbrock formula

Trapezoidal rule using a

ode23t "free" interpolant

(Shampine et al., 1999)

TR-BDF2
ode23tb


(Shampine and Hosea, 1996)

Runge-Kutta,

ode45 Dormand-Prince (4,5) pair

(Dormand and Prince, 1980)

Chapter 2. Biped Model and Simulator: Configuration, Visualisation and Accuracy 66

One-step method refers to the need of only the solution at the preceding time point, whereas

multistep method refers to the need of solutions at several preceding time points in order to

compute the current solution.

Explicit one-step method involves the calculation of the system’s state at the next time step

by using the current system state, and can be defined as

(' + ∆') = ( ('))

(2.16)

where (') is the current system state, (' + ∆') is the state at the next time step, and 

indicates a function.

Implicit one-step method involves the calculation of the system’s state at the next time step

by using both the current system state and the next system’s state, and can be obtained by

solving for the function V in

V( ('), (' + ∆')) = 0

(2.17)

where (' + ∆') is the required solution.

The implicit method requires extra computation, but is optimal in solving stiff problems,

while the explicit method takes little computation time but are not efficient in dealing with

stiff systems.
Chapter 2. Biped Model and Simulator: Configuration, Visualisation and Accuracy 67

2.5.2 Causes and effects of stiffness in models with rapid dynamics

Stiffness is a concept in the numerical solution of differential equations (Spijker, 1996). An

ODE problem is stiff if the solution being sought varies slowly but nearby solutions change

rapidly, hence requiring the simulation step size to be reduced down to unacceptably small

values. For our research work, we are mainly focusing on the efficiency aspect for

computational time. While explicit solvers will solve stiff problems, they take too long to do

them because the step size is reduced to extremely small values. Hence, there is a need to

select a solver that will help our simulations to complete at a faster pace at reasonable

accuracies, given that there are situations where our biped model will encounter stiff

situations. This selection of ODE solver is one of the most essential steps when creating the

simulation world for our biped model.

To understand what stiffness is and how it will affect the simulation speed, we look at one

example, which is that of a simple combustion model (Abelman and Patidar, 2008) and whose

dynamics is similar, in terms of stiffness, to the reaction force dynamics obtained in our biped

model during ground surface contact. We will not go into the details of linear stability and

eigenvalues of the Jacobian equations, but only look at the computational time aspect in

using different ODE solvers. In the combustion model, when one lights up a match, the ball of

flame grows rapidly until it reaches a critical size, where it remains because the amount of

oxygen consumed by the combustion in the interior of the ball balances the amount available

through the surface.

Chapter 2. Biped Model and Simulator: Configuration, Visualisation and Accuracy 68

This simple model is defined by the equations

W#
= #! − #3
W' (2.18)

#(0) = X (2.19)

2
0<#<
X
(2.20)

where the scalar variable #(') represents the radius of the ball and the # ! and # 3 terms

comes from the surface and area of the volume. X is the critical parameter and is the initial

radius. A solution over a length of time that is inversely proportional to X is required.

With respect to our biped model, the simulation step size will be reduced to very small values

whenever the biped model is about to touch the ground surface, and which could lead to

stiffness in the system. Having a step size which is too high will make the model go below the

ground surface by the next simulation step, due to the force of gravity. Having too low a

value for the step size will adversely affect the speed of simulation. However, while reducing

the step size will likely improve the accuracy of the simulation, the simulation speed and real

time taken will be negatively affected, as indicated through Figure 2.28 for the simple

combustion model. The combustion model is being used here for its simplicity in getting to

understand the relationship between simulation step sizes, ODE solvers, and computational

time taken. The usage of the actual biped model will be much more complicated and will be

further explored in subsection 4.4.4.

Chapter 2. Biped Model and Simulator: Configuration, Visualisation and Accuracy 69

Varying the δ value in ode45 (Explicit one-step method)

Absolute tolerance = 1e- 4; Relative tolerance = 1e- 4

δ = 0.1, Real time taken = 0.01 s Magnified plot for δ = 0.1 near critical point
1 1.0002

0.5 1

0 0.9998
0 5 10 15 20 10 10.5 11

δ = 0.01, Real time taken = 0.05 s Magnified plot for δ = 0.01 near critical point
1.5 1.0002

1
1
0.5

0 0.9998
0 50 100 150 200 100 105 110

δ = 0.001, Real time taken = 0.33 s Magnified plot for δ = 0.001 near critical point
1.5 1.0002

1
y 1
0.5

0 0.9998
0 500 1000 1500 2000 980 1000 1020 1040 1060 1080 1100 1120

δ = 0.0001, Real time taken = 3.05 s Magnified plot for δ = 0.0001 near critical point
1.5 1.0002

1
1
0.5

0 0.9998
0 0.5 1 1.5 2 0.98 1 1.02 1.04 1.06 1.08 1.1 1.12
4 4
x 10 x 10
δ = 0.00001, Real time taken = 28.03 s Magnified plot for δ = 0.00001 near critical point
1.5 1.0002

1
1
0.5

0 0.9998
0 0.5 1 1.5 2 0.98 1 1.02 1.04 1.06 1.08 1.1 1.12
5 5
x 10 x 10

δ intervals

Figure 2.28: X analysis. Reducing the simulation step size increases the accuracy of the results at the
expense of computational time taken
Chapter 2. Biped Model and Simulator: Configuration, Visualisation and Accuracy 70

The real time taken for different values of X is now examined from Figure 2.28. The ODE

solver used is an explicit variable-step one not optimised for stiff problems, but which

provides good accuracy (Shampine, 1994), as mentioned in Table 2.13. It is called the ode45

solver, which makes used of the Runge-Kutta and Dormand-Prince (4,5) pair. The absolute

and relative tolerance values are set to 1e-4, meaning that the computed state is accurate to

within 0.01%, with any error that exceeds this tolerance value leading to a reduction in the

solver’s time step size and taking additional computational time for a solution satisfying the

tolerances.

Table 2.13: Comments on variable-step ode solvers within Simulink

ODE Order of
Comments
Solver Accuracy
ode113 can be more efficient than ode45 at stringent tolerances or for Variable, Low
ode113
computationally intensive problems. to High

Numerical differentiation formulas are related to, but are more efficient
Variable, Low
ode15s than the backward differentiation formulas, which are also known as Gear's
to Medium
method. ode15s is efficient for stiff problems.

ode23 is more efficient than ode45 at crude tolerances and in the presence
ode23 Low
of mild stiffness.

ode23s is more efficient than ode15s at crude tolerances, and can solve
ode23s Low
stiff problems for which ode15s is ineffective.

ode23t is used if the problem is only moderately stiff and a solution with no
ode23t Low
numerical damping is needed.

ode23tb is more efficient than ode15s at crude tolerances, and can solve
ode23tb Low
stiff problems for which ode15s is ineffective.

Generally, ode45 is the best to apply as a first try for most problems. ode45
ode45 is more accurate and faster than ode23. If ode45 is slow computationally, Medium
the problem may be stiff and thus in need of an implicit solver.
Chapter 2. Biped Model and Simulator: Configuration, Visualisation and Accuracy 71

Initially, the problem is not stiff, but becomes stiff as the solution approaches steady state.

Unlike at the beginning, when only a few steps were required to plot the value of 0, once the

rapidly changing solution is encountered, the ODE solver starts to compute significantly more

steps in order to be able to plot the value of 1 within the tolerance value.

Our biped model will encounter similar situations, at the instant when parts of the biped are

about to touch the ground surface. While above the ground surface, no normal reaction force

will be encountered by the body parts, but once the ground surface gets into contact, the

normal reaction force will rapidly change to another value, and thereby potentially making

our biped system stiff. This stiffness has a significant role in our biped system, particularly

with regards to the simulation speed, because the normal reaction force will be used by the

ODE solver in order to obtain positions and velocities of the rigid bodies within the system.
Chapter 2. Biped Model and Simulator: Configuration, Visualisation and Accuracy 72

2.5.3 Implicit and explicit ODE solvers for the biped model

After looking at how the simulation time step affects the computational time and overall

accuracy of solutions, the remaining variable-step ode solvers from Table 2.13 are considered

to see how well they deal with stiff problems. Continuing with the flame combustion model,

due to its similarity to the biped model but with the advantage of being easier to understand

and get conclusions from, X is set to a fixed value of 0.0001 for its reasonable results in Figure

2.28. The absolute and relative tolerances are set to be either 1e-4 or 1e-10, but both types of

tolerance have the same value during any one simulation. The results for explicit ODE solvers

are shown in Figure 2.29, and those for implicit ODE solvers are presented in Figure 2.30.

Using different explicit ODE solvers with δ = 0.0001

Absolute tolerance and relative tolerance set to the same value

Full plot for chosen Magnified plot with Magnified plot with
explicit ODE solver both tolerances = 1e- 4 both tolerances = 1e- 10

Number of steps = 5878 Number of steps = 6573

ode113 (Multistep method) Real time taken = 2.16 s Real time taken = 2.41 s
1.5 1.0002 1.0002

1
1 1
0.5

0 0.9998 0.9998
0 1 2 1 1.05 1.1 1 1.05 1.1
4 4 4
x 10 x 10 x 10

Number of steps = 3955 Number of steps = 8245

ode23 (One-step method) Real time taken = 1.80 s Real time taken = 3.75 s
1.5 1.0002 1.0002
y
1
1 1
0.5

0 0.9998 0.9998
0 1 2 1 1.05 1.1 1 1.05 1.1
4 4 4
x 10 x 10 x 10

Number of steps = 12153 Number of steps = 13129

ode45 (One-step method) Real time taken = 3.04 s Real time taken = 3.15 s
1.5 1.0002 1.0002

1
1 1
0.5

0 0.9998 0.9998
0 1 2 1 1.05 1.1 1 1.05 1.1
4 4 4
x 10 x 10 x 10
δ intervals

Figure 2.29: Explicit ODE solvers. Effect of tolerances for different explicit variable-step ODE solvers
Chapter 2. Biped Model and Simulator: Configuration, Visualisation and Accuracy 73

Using different implicit ODE solvers with δ = 0.0001

Absolute tolerance and relative tolerance set to the same value

Full plot for chosen Magnified plot with Magnified plot with
implicit ODE solver both tolerances = 1e- 4 both tolerances = 1e- 10

Number of steps = 90 Number of steps = 713

ode15s (Multistep method) Real time taken = 0.05 s Real time taken = 0.27 s
1.5 1.0002 1.0002

1
1 1
0.5

0 0.9998 0.9998
0 1 2 1 1.05 1.1 1 1.05 1.1
4 4 4
x 10 x 10 x 10

Number of steps = 69 Number of steps = 4953

ode23s (One-step method) Real time taken = 0.05 s Real time taken = 3.45 s
1.5 1.0002 1.0002

1
1 1
0.5

0 0.9998 0.9998
0 1 2 1 1.05 1.1 1 1.05 1.1
y 4 4 4
x 10 x 10 x 10

Number of steps = 87 Number of steps = 7228

ode23t (One-step method) Real time taken = 0.05 s Real time taken = 1.75 s
1.5 1.0002 1.0002

1
1 1
0.5

0 0.9998 0.9998
0 1 2 1 1.05 1.1 1 1.05 1.1
4 4 4
x 10 x 10 x 10

Number of steps = 75 Number of steps = 5735

ode23tb (One-step method) Real time taken = 0.06 s Real time taken = 2.93 s
1.5 1.0002 1.0002

1
1 1
0.5

0 0.9998 0.9998
0 1 2 1 1.05 1.1 1 1.05 1.1
4 4 4
x 10 x 10 x 10
δ intervals

Figure 2.30: Implicit ODE solvers. Effect of tolerances for different implicit variable-step ODE solvers
Chapter 2. Biped Model and Simulator: Configuration, Visualisation and Accuracy 74

Based on the plots obtained in Figure 2.29 and Figure 2.30, a summary of the real time taken

and number of steps present in the solutions are shown Figure 2.31. It should be noted that

running the same simulation multiple times will lead to negligible differences in the real time

taken but the number of steps taken to obtain the solution while using a specific ODE solver

will remain unchanged.

Time taken for solution for a stiff Time taken for solution for a stiff
problem when both tolerances = 1e-4 problem when both tolerances = 1e-10

Explicit ODE Solver Implicit ODE Solver Explicit ODE Solver Implicit ODE Solver

ode113 ode113
ode23 ode23
ODE solver

ODE solver
ode45 ode45
ode15s ode15s
ode23s ode23s
ode23t ode23t
ode23tb ode23tb
0 1 2 3 4 0 1 2 3 4
Time taken (s) Time taken (s)

(a) (b)
Steps taken for solution for a stiff Steps taken for solution for a stiff
problem when both tolerances = 1e-4 problem when both tolerances = 1e-10

Explicit ODE Solver Implicit ODE Solver Explicit ODE Solver Implicit ODE Solver

ode113 ode113
ode23 ode23
ODE solver

ODE solver

ode45 ode45
ode15s ode15s
ode23s ode23s
ode23t ode23t
ode23tb ode23tb
0 5,000 10,000 15,000 0 5,000 10,000 15,000
Number of steps Number of steps

(c) (d)

Figure 2.31: Summary of time taken and number of steps for different ODE solvers (a) The real time
taken when the absolute and relative tolerances are set to 1e-4 (b) The real time taken when the
absolute and relative tolerances are set to 1e-10 (c) Number of steps in the solution when the absolute
and relative tolerances are set to 1e-4 (d) Number of steps in the solution when the absolute and
relative tolerances are set to 1e-10
Chapter 2. Biped Model and Simulator: Configuration, Visualisation and Accuracy 75

Based on Figure 2.29, Figure 2.30, and Figure 2.31, the following conclusions could be

obtained regarding the ODE solvers:

(1) ode15s could be an appropriate ODE solver given it takes very few steps and less time

in finding solutions, for both crude and stringent tolerance values.

(2) ode113 does not provide stable accuracy at crude tolerances.

(3) Compared to explicit ODE solvers, implicit ODE solvers lose a significant amount of

time advantage at very stringent tolerance values, with the exception of the ode15s.

(4) Given that a tolerance value close to 1e-4 is good enough for our biped model, implicit

ODE solvers seems to be the better choice.

(5) Among explicit ODE solvers, ode23 is the better choice at crude tolerance values in

terms of real time taken.

(6) With the highest amount of steps present for a solution by using ode45 in the tested

cases, the latter is likely to be more accurate than the other ODE solvers.

(7) At crude tolerance values, ode45 is less efficient than the other explicit ODE solvers in

terms of real time taken. However, it is the only one among the explicit ODE solvers

in being able to lie within the tolerance limits set near the rapidly changing solution.

(8) Overall, the explicit ODE solvers are primarily concerned with accuracy. They reduce

the step sizes whenever stiff problems are encountered, and unfortunately making

them less time efficient in the process.

(9) Overall, implicit solvers are less concerned about efficient integration, favouring

larger step sizes while retaining numerical stability. Hence, for crude tolerances, they

tend to lose accuracy while gaining significant advantages in terms of real time taken

in order to reach a solution.

Chapter 2. Biped Model and Simulator: Configuration, Visualisation and Accuracy 76

Using the conclusions obtained from this section on the selection of an appropriate ODE

solver for the biped model, it appears that implicit ODE solvers would be more efficient in

terms of real time taken, but later analysis while using the actual biped model does not give

such a straightforward conclusion. However, with a better understanding of the ODE solvers

available within the SimMechanics environment, and a basic idea about stiff and nonstiff ODE

solvers, the conclusions in terms of speed and accuracy for the biped model will be easier to

follow. The analysis for the biped model is covered in subsection 4.4.4, where the horizontal

ground surface modelling and other parameters within the SimMechanics environment have

significant effects on the performance and accuracy of the different ODE solvers.
Chapter 2. Biped Model and Simulator: Configuration, Visualisation and Accuracy 77

2.6 Conclusion

This chapter presented all the essential components involved in the development of a fully

functional model within the SimMechanics software. Configurations and procedures chosen

for the creation of the mobile robot simulator and biped model were explained, and the

different methods of visualising SimMechanics models were covered. Conclusions from the

computational aspects of the different ODE solvers within Simulink were obtained, and also

with regards to the presence of stiffness in the biped model. Based on the understanding of

the ODE solvers covered in this chapter, appropriate settings and tolerances for the ODE

solvers will be made during the simulation of the biped model, and depending on the need

for accuracy or simulation speed.

Combining all the configurations, steps and explanations covered in this chapter, the fully

functional biped model can be created. Taking one of the simplest subsystems in the biped

model, with only two actuated joints, as an example, a fully functional neck and head

subsystem of the biped model will have the structure shown in Figure 2.32. As an indication,

the whole biped model will consist of 28 actuated joints in total, and hence it is easier to

learn how to make just a subsystem fully functional before dealing with the complete biped

model in SimMechanics. In conclusion, Figure 2.32 acts as the summary to most of the

sections of this chapter.

Chapter 2. Biped Model and Simulator: Configuration, Visualisation and Accuracy 78

Figure 2.32: A fully functional subsystem of the biped model. Combining all the configurations, steps
and explanations covered in this chapter, a fully functional biped model, consisting of fully functional
subsystems, within the SimMechanics environment will be realisable