Proceedings of the 2009 IEEE

International Conference on Robotics and Biomimetics

December 19 -23, 2009, Guilin, China

Dynamic Stability Gait Planning of kid Humanoid Robot

Jie Zhao, Le Qi, Jihong Yan , Yanhe Zhu
State Key Laboratory of Robotic Technology and System, Harbin Institute of Technology,
Harbin, Heilongjiang Province, 150001. China
Abstract - We adopt multi-point planning method based on optimization problems in a complex system. It does not rely
cubic spline function to perform parametric design of dynamic on the specific field. With strong applicability, it can be
walking. The objective of optimization is to achieve maximum used in various types of problems and wide range of
ZMP stability and minimum joint driving torque, together subjects[3].
with the combination of multi-objective genetic algorithm to The second part of the article is the robot kinematics
optimize the gait parameters. Simulation verification has also modeling and the solution of obverse and inverse
been carried out in ADAMS and Matlab after optimization. In kinematics. The third part adopts the multi-point planning
addition, we tested dynamic walking on kid humanoid robot method based on cubic spline function to carry out a
platform, by which the realization of dynamic and steady parametric design on dynamic walking and simple
walking of physical prototype further proves the validity of the humanoid action, together with genetic algorithm to
theoretical research. Moreover, weve made a preliminary optimize the above parameter. Part four and part five are the
study on the planning method of performing complex actions simulation and the prototype experiments of the robot as
such as push-ups, by which the realization of scheduled actions well as the analysis of the results.
through simulation and physical prototype experiments also II. KINEMATICS MODELING OF BIPED ROBOT
proved the rationality of the planning. The height of the robot is 350mm, the weight is about
Index Terms - genetic algorithm; ZMP; simulation; gait 2.3kg and the degree of freedom is 17.
planning Kinematics modeling is designed to describe the
I. INTRODUCTION relationship between the position, posture and the joint
Robots are the consequence of the development of angle of a robot. There are many robot kinematics modeling
modern science and technology. The supreme goal is to methods[4]. The article, through adopting the modeling
create robots like human beings to walk, work and method based on generalized coordinates, is to realize
communicate with people. kinematics modeling of humanoid robot.
Gait planning is one of the most important A. Kinematics Modeling of Dynamic Walking
technologies of humanoid robot, which is not only a 1) Forward Kinematics Modeling:
critical technology for the biped robot to realize the most
basic humanoid action, but also serves as a theoretical 3
guidance and technical support for mechanism design
and optimization of the humanoid robot[1]. Without gait
planning, the basic movement of the robot can never be
2
realized, let alone the advantages of flexibility and
4
adaptability, and other functions of the humanoid robot.
1
Previous researches are always based on experience and 5
intuition, which is not only laborious and time-consuming
but also difficult to obtain optimized results[2]. Genetic
algorithm provides a common framework to solve
Fig.1 Model of Front plane

978-1-4244-4775-6/09/$25.00 2009 IEEE. 2373

 We simplify the robot to five connecting rods models in
the forward plane, as shown in Figure 1.
Let the left foot of the robot be the support foot, then
7 9
6 8
establish the origin O(0,0) of the reference coordinate
system at the left ankle joint. Let the coordinate of each
joint in the Cartesian coordinate system be (xi, zi), the
angle of each joint relative to axis Z be i , the actual
Fig.2 Model of Side plane and Side angle
angle relative to itself be i , the length of each
right foot be the swing foot, the lateral model is required to
connecting rod be li , the quality of the connecting rod
meet constraints at any time:
be mi , the coordinate of the rod centroid in the
Cartesian coordinate system be ( xci , zci ) , the distance 6 = 7 = 8 = 9 = ce
from the rod centroid to the joint axis below be d i . It is
The calculation ce is shown in figure 3, for dynamic
easy to obtain the coordinate of each joint in the
Cartesian coordinate system. Let the angle of i which walking, the projection of gravity center will not always fall

is on the left positive direction of axis Z be positive, right on the support plane of the support foot, therefore, let

be negative. ce be the parameters, and restrict its value within range

of ce arctan ( a / b ) and finally determine the value
Since we make constraint to keep the upper body and the
through optimization.
ground vertical during gait planning, thus we get the result
B. Kinematics Modeling of Complex Actions
3 0 (0-1).
Through the research, push-ups and roll actions
Let the coordinate of hip be ( xk , zk ) , the coordinate of the
(including pre-roll and post-roll) have two characteristics:
swing leg be ( xh , zh ) , due to the multiple solutions of
on one hand, these actions are basically achieved by
inverse kinematics, we restrain the scope of joint angle to
forward joints; on the other hand, the postures of both right
get the required expression. The following are the results
and left sides of the limbs are symmetrical. Therefore,
through calculating:
simplify the biped robot to four connecting rods models in
x + z +l l
2 2 2 2
xk the forward plane, and the modeling method is similar to
1 = arccos k k 1 2
+ arctan
2l1 x + z 2
k
2
k
zk dynamic walking.
III. MOVEMENT PLANNING
xk l1 sin 1
2 = arctan A. Gait Planning
zk l1 cos 1
xh xk l5 sin 5 In order to overcome the previous deficiencies of
4 = arctan
zk zh l5 cos 5 three-point planning, more points are inserted among three
key points. In addition, we restrict the posture of the swing
( xh xk ) 2 + ( zk z h ) 2 + l52 l42 x xk
5 = arccos + arctan h foot, the speed and acceleration of landing. Through careful
2l5 ( xh xk ) 2 + ( z k zh )2 zk zh
study, we found that as long as two key points after starting
2) Lateral Kinematics Modeling point and before falling point are added, the smoothness of
trajectory can be ensured.[5]
The lateral modeling of the humanoid robot is shown in
As shown in Figure 3, the concept of five-point planning
figure 2:
is that add b) and d) based on the three-point a), c) and e).
During gait planning, we set the robot to keep its upper Through setting the time interval between points a) and b)
body perpendicular to the ground, to keep the robot's swing to achieve the effect of retarding lift of the swing foot from
foot parallel with the ground. Therefore, when we let the the ground so as to reduce the collision force when the

2374
interference occurred between the ground and the swing
L
foot; Through setting the time interval between points d) kL 2 ...........................t = kT
and e) to retard the speed of the swing foot when landing,
kL L + l .....................t = kT + t
thus, the collision force can be reduced between the ground 2
m m

and the swing foot, thereby the goal of buffer could be xh ( t ) = kL..................................t = kT + th
L
achieved as well; Through setting the position of points b) kL + ln .....................t = kT + tn
and e) in the movement space, to prevent the interference 2
L
between lifting and landing of the swing foot and the kL + ...........................t kT + TS , ( k + 1) T
2
ground. The insertion of points b) and d) not only
compensate the deficiencies of three-point dynamic
0....................................t = kT
planning, but also guarantee the smoothness of the
hm ..................................t = kT + tm
trajectory. Therefore, five-point planning is a more
zh ( t ) = H ..................................t = kT + th
reasonable and effective method to plan dynamic walking. h ..................................t = kT + t
n n

0...................................t kT + TS , ( k + 1) T

In the above equations, k is an integer.

The initial velocity of the swing foot is zero at the
moment of kT, and the velocity of landing is expected to be
Fig.3 Five points gait planning
zero as well. Then, the impact between the foot and the
At first, work out the trajectories of the swing foot and ground will be the minimum. Therefore, let the derivative
hip on the positive plane, then, plan the lateral joint on the of the swing foot at the moment kT, kT+Ts and xh(t), zh(t) be
side plane, finally, calculate the trajectory of each joint zero.
through geometric relationship. The cubic spline function could be used to calculate
1) Planning of the Leg Trajectory the relationship between the coordinates X, Y of the
Within a walking cycle T, let TS and TD be the period of ankle joint and the time t respectively. Through the cubic
single-foot support and bipedal support respectively. spline formula and the second-order central difference
Dynamic walking requires the robot a high speed walking formula we can get the results.
without long time of bipedal support. Let the time for The value can be obtained by utilizing Gauss
bipedal support occupies 20% of the entire walking cycle. elimination method, thereby the expression of xh(t) and
Let the height of lifting be H , the length of pace be L, a zh(t) in the cycle T. We will not deduce here.
foot long. Let the initial time of the pace k be kT , then, 2) Planning of the Hip Trajectory
the time to complete pace K is k (T + 1) . The planning of the hip we adopt, however, is similar to
Let the left foot be the swing foot, the origin of the the planning of foot. As long as ensuring a uniform
coordinate in XOZ plane be the center of the support foot, movement of the hip joint in the direction X is all right.
the initial coordinate of cycle k is (kL, 0) , the top position Therefore, only three points are required to plan the hip
of the swing foot is (kL + L / 2, H ) , the coordinate of trajectory in the direction X. The following is the constraint
landing is ((k+1)L,0). We adopt five-point planning method. equation of the hip joint:
Apart from the above three points, two more points In the equations, k is an integer, le is the distance from
(kL + lm , hm ) and (kL + ln , hn ) can be inserted before lifting the initial moment of the hip joint to the origin coordinate in
and landing of the swing foot. Through the above five the direction X.
interpolation points we can obtain the following constraint
equations:

2375
 kL le ..................t = kT The relationship between ce and the time t can be

T obtained through utilizing cubic spline function. Due to the
xk ( t ) = kL........................t = kT +
2 similar solution as swing foot and hip joint, we will not
kL + le ..................t = ( k + 1) T repeat it here.
Likewise, through utilizing cubic spline function we can 4) Trajectory Planning of Other Joints
work out the relationship between coordinate X of the hip In the previous section weve planned the trajectory of
joint and the time t. For the concrete calculation steps you foot ( x ( t ) , z ( t ) ) and the hip ( x ( t ) , z ( t ) ) in addition,
h h k k
can refer to the planning of swing foot.
we obtained 3 ( t ) by the posture planning of the upper
The following is the plan of the hip trajectory in the
body. Through the inverse kinematics computing we can
direction Z. Since the quality of the robot is mainly
work out the angle of each rod in the forward plane so as to
concentrated in the upper part of the body, we may
get further result of all the joint angles.
approximately take the position coordinate Z of the hip joint
From the inverse kinematics computing we obtain:
as the gravity position of coordinate Z. When the robot is
walking, if the height between the gravity center and the xk2 ( t ) + zk2 ( t ) + l12 l22 xk ( t )
1 ( t ) = + arctan
ground can be kept at a fixed value, the inertial force in the 2l1 xk2 ( t ) + zk2 ( t ) zk ( t )
vertical direction and the impact between the ground and
the landing swing foot could be reduced. Therefore, let the xk ( t ) l1 sin 1
2 ( t ) = arctan
coordinate of the hip in the direction Z be a constant zk ( t ) l1 cos 1
value H k , then we obtained:
xh ( t ) xk ( t ) l5 sin 5
zk ( t ) = H k t kT , ( k + 1) T 4 ( t ) = arctan
zk ( t ) zh ( t ) l5 cos 5
The value of H k can be determined by the formula. The
( x (t ) x (t )) + ( z ( t ) z (t )) + l l
2 2 2 2
given pace length L can be obtained through the geometric
5 ( t ) =
h k k h 5 4

2l ( x ( t ) x ( t ) ) + ( z ( t ) z ( t ) )
2 2
relationship of the rods.
5 h k k h

( l1 + l2 )
2
L2 H k l1 + l2
xh ( t ) xk ( t )
+ arctan
The above is the plan for the trajectory of the hip joint. zk ( t ) zh ( t )
3) Planning of Lateral Plane
B. Planning of Complex Actions
The lateral joints of the robot satisfied the
We can take the planning method of walking gait as a
formula 6 = 7 = 8 = 9 = ce .We can conclude that reference. Firstly, plan several key postures of the whole set
the planning of the lateral plane movement for the biped of actions, and then utilizing cubic spline interpolation for
robot is the planning of ce . The function of the lateral smooth and constant track, and finally obtain the
joints is to convert the gravity center between the two legs. relationship among the time and all the angles applied to
The entire walking is mainly completed in the period of control the joint through inverse kinematics.
single-foot support. The formula (2-6) ce should satisfy
the following constraints:

C.................t = KT

ce (t ) = 0..................t = t f
Fig. 4 key postures of push-up
C...............t = K (T + 1)
As shown in Figure 4, we can decompose the push-up
In the above formula, C arctan ( a / b )

2376
action into several key postures: in the course of movement. We select a group of parameters
According to the above key postures, the position of the randomly in the variable range to simulate and then make
hip joint and the hand on plane XOZ could be designed comparison between the simulation results and the results
through the geometric relationship of all the robs. after optimization. We take forward joint as an example
Cubic spline function could be used to calculate the here, Figure 6 shows the torque curve of forward joint
interpolation of all the above points so as to generate before optimization through simulation; the torque curve of
constant trajectory of the hip and the ankle joints. Then, forward joint after optimization is shown in Figure 7.
substitute it into the equation of the coordinate system,
constant curve of each joint angle and time can be obtained.
Roll movements (including pre-roll and post-roll) are
similar to push-ups because they are all being completed in
forward plane, and the actions of both right and left limbs
Fig.6 torque curve of forward joint before optimization
are symmetrical. Therefore, the planning is also similar to
the method of planning push-ups, thus we are not going to
introduce it again.
Finally, the gait will be optimized through genetic
algorithm. The ZMP trajectory after optimization will be
closer to the center of support field. Thereby it enables the Figure.7 torque curve of forward joint after optimization
robot to obtain a greater degree of stability margin, and The comparison demonstrates that the peak torque of the
further guarantees the stability of dynamic walking. forward joint and the lateral joint after optimization is
Meanwhile, the goal of expected optimization has been obviously less than the torque before optimization. It can be
reached as well. seen as well that in the simulation animation the stability of
IV. SIMULATION AND EXPERIMENTS gait after optimization is superior to the optimization before.
A. Simulation of Dynamic Walking Therefore, the dynamics simulation results prove a high
We import the joint angle after optimization and time stability of gait and a small drive of joints after optimization.
curve into ADAMS to drive each joint of the robot. It has already met the requirements of the target and thus
Through the adjustment of some parameters in ADAMS, can be used to test the prototype.
the robot realizes a fast and smooth walking as shown in B. Simulation of Complex Actions
Figure 5.

Fig.8 Key frames of push-up action

Fig.5 walking key frames of positive plane

and walking key frames of lateral plane

Through post-processing function, analysis can be

performed after obtaining each joint torque curve of the leg
Fig.9 Curve of time and angle of each joint on the left

2377
 As shown in Figure 8, import the generated data into The subject makes a research on gait planning on the
ADAMS to make dynamics simulation, and we get the basis of small humanoid robot. According to generalized
complete push-up actions finally through adjustment. The coordinates, the robot kinematics model has been established,
curve of time and angle of each joint on the left of the robot and the solution of obverse and inverse kinematics has been
are shown in Figure 9. Pre-roll and post-roll are similar to worked out as well. Based on the method of ZMP stability
this, thus, we will not discuss the simulation result here. criterion, the planning of dynamic walking has accomplished.
In addition, we adopt the five-point planning method of cubic
C. The Prototype Experiments
spline function to perform parametric design of dynamic
The simulation results of Matlab and ADAMS prove that
walking as well as optimizing the gait parameters through
the planning gait could accomplish dynamic stability
genetic algorithm, thereby obtaining an optimized walking
walking of humanoid robot. Therefore, the experiments
data with relatively high stability. The comparison results of
performed of small humanoid robot on the dynamic
dynamics simulation show that multi-point planning method
walking test platform achieve dynamic walking at the speed
is suitable for planning dynamic walking of humanoid robot.
of 74mm/s. The prototype key postures of dynamic walking
The experimental results indicate that the robot is able to walk
are shown in Figure 10.
fast and smooth at a speed of 74mm/s. Moreover, we make a
preliminary research on the planning of complex actions, such
as push-ups and rolls. The simulation experiments have been
accomplished as well. In addition, the realization of push-ups
during prototype experiments proved the rationality of the
plan.
REFERENCES
[1] Y. Sakagami, R. Watanabe, C. Aoyama, S. Matsunaga, N. Higaki, and K.
Fujimura, "The intelligent ASIMO: System overview and integration," Proc.
Fig.10 Key gait of dynamic walking IEEE/RSJ Int. Conference on Intelligent Robots and Systems, pp. 2478-2483,

Since the mechanical body is not perfect at present, 2002.

weve only completed the prototype experiment of push-ups. [2] "Special Issue: The Final Reports of HRP and Perspectives of Humanoid

The experiment method is the same as dynamic walking. Robot Developments," Journal of the Robotics Society of Japan, Vol.22,

Through online adjustment, the accomplishment of the No.1, pp. 1- 54, 2004.

push-up actions of physical prototype proved the [3] Kweon Soo Jeon; Ohung Kwon, Jong Hyeon Park, " Optimal trajectory

correctness of the planning. The key postures of push-up are generation for a biped robot walking a staircase based on genetic algorithms"

shown in Figure 11. Proceedings. 2004 IEEE/RSJ International Conference on

[4] Jae Won Kho; Dong Cheol Lim; Tae Yong Kuc, "Implementation of an
Intelligent Controller for Biped Walking Robot using Genetic Algorithm"
Industrial Electronics, 2006 IEEE International Symposium on
[5] C. Zhu, A. Kawamura, "Walking Principle Analysis for Biped Robot with
ZMP Concept, Friction Constraint, and Inverted Pendulum Model,"
Proceedings of the 2003 IEEE/RSJ IROS, Las Vegas, Nevada, USA, pp.
364-369, October 2003.

Fig.11 Key gait of push-up

V. CONCLUSION

2378