Today

Robotics and Autonomous Systems

Lecture 2: Mobile Robotics

Richard Williams
• Today we’ll start to look at the main problems faced by mobile robots.
• This sets up the issues we’ll consider for the first half of the course.
Department of Computer Science
University of Liverpool • We’ll also consider how these issues relate to the idea of agency.

1 / 48 2 / 48

Autonomy Autonomy

• A really autonomous vehicle makes its own decisions about what to

do.
• Many autonomous vehicles are not really autonomous percepts

Environment sensors

effectors
• They are teleoperated.
actions
• The notion of an agent can help us understand what this requires.

3 / 48 4 / 48
What is an agent? What is an agent?

• Trivial (non-interesting) agents:

• As we said before: • thermostat;

• light switch;
An agent is a computer system that is situated in some
• unix daemon (e.g., biff).
environment, and that is capable of autonomous action in
that environment in order to meet its delegated objectives.

• It is all about decisions

• An agent has to choose what action to perform.
• An agent has to decide when to perform an action.

• More interesting agents are intelligent.

5 / 48 6 / 48

Intelligent Agents Abstract Architectures for Agents

• Assume the world may be in any of a finite set E of discrete,

instantaneous states:
E “ te , e 1 , . . .u.
• An intelligent agent is a computer system capable of flexible
autonomous action in some environment. • Agents are assumed to have a repertoire of possible actions available
By flexible, we mean: to them, which transform the state of the world.
• reactive;
• pro-active;
Ac “ tα, α1 , . . .u
• social. • Actions can be non-deterministic, but only one state ever results from
• All these properties make it able to respond to what is around it. an action.
• A run, r, of an agent in an environment is a sequence of interleaved
world states and actions:

α
0 1 α2 3 α α αu´1
r : e0 ÝÑ e1 ÝÑ e2 ÝÑ e3 ÝÑ ¨ ¨ ¨ ÝÑ eu

7 / 48 8 / 48
Runs of agents Runs of agents

• When actions are deterministic each state has only one possible
successor.
• A run would look something like the following:

9 / 48 10 / 48

Runs of agents Runs of agents

North!

11 / 48 12 / 48
Runs of agents Runs of agents

North!

13 / 48 14 / 48

Runs of agents Runs of agents

East!

15 / 48 16 / 48
Runs of agents Runs of agents

North!

17 / 48 18 / 48

Runs of agents Runs of agents

• When actions are non-deterministic, a run (or trajectory) is the same,

• Which we might picture as so: but the set of possible runs is more complex.

North!
North!
North!
North!

19 / 48 20 / 48
Runs of agents Agents

• In fact it is more complex still, because all of the runs we pictured • We can think of an agent as being a function which maps runs to
start from the same state. actions:
• Let R be the set of all such possible finite sequences (over E and Ac)
This is the set of all runs from all starting states. Ag : RE Ñ Ac
• RE is the subset of R that end with a state. • Thus an agent makes a decision about what action to perform based
• All the ones where the agent needs to make a decision. on the history of the system that it has witnessed to date.

21 / 48 22 / 48

Agents Purely Reactive Agents

North! • Some agents decide what to do without reference to their history —

they base their decision making entirely on the present, with no
North! North! reference at all to the past.
West! • We call such agents purely reactive:

action : E Ñ Ac

• A thermostat is a purely reactive agent.

"
off if e = temperature OK
actionpe q “
• Potentially the agent will reach a different decision when it reaches on otherwise.
the same state by different routes.

23 / 48 24 / 48
Purely Reactive Agents Purely Reactive Agents

North!

North! North!

West!

• A reactive agent will always do the same thing in the same state.

25 / 48 26 / 48

A reactive robot Agents with state

• A simple reactive program for a robot might be:

Drive forward until you bump into something. Then, turn to percepts

the right. Repeat.

see state

Environment

next
action

actions

27 / 48 28 / 48
Agents with state Agents with state

• The action-selection function action is now defined as a mapping

• The see function is the agent’s ability to observe its environment,
whereas the action function represents the agent’s decision making action : I Ñ Ac
process.
• Output of the see function is a percept: from internal states to actions.
• An additional function next is introduced, which maps an internal
see : E Ñ Per state and percept to an internal state:
• The agent has some internal data structure, which is typically used to next : I ˆ Per Ñ I
record information about the environment state and history.
• Let I be the set of all internal states of the agent. • This says how the agent updates its view of the world when it gets a
new percept.

29 / 48 30 / 48

Agents with state A robot with state

percepts

see state

Environment

1 Agent starts in some initial internal state i0 . next

action

2 Observes its environment state e, and generates a percept see pe q. actions

3 Internal state of the agent is then updated via next function,

becoming next pi0 , see pe qq.
• per is a bool that indicates “against an object”.
4 The action selected by the agent is actionpnext pi0 , see pe qqq.
• i is an integer, “against object for n steps”.
This action is then performed.
• see updates per each step, indicating if the robot is against an object.
5 Goto (2).
• next is as follows:
"
i+1 if per = true
next pi q “
0 otherwise.

31 / 48 32 / 48
A robot with state What is mobile robotics?

• Last time we boiled the challenges of mobile robotics down to:

?
• Now the robot can take more sophisticated action.
• For example, backing up if it cannot turn away from the wall
immediately.

• This is an example of a common situation in robotics.

• Trading memory and computation for sensing. • Where am I ?
• Where am I going ?
• How do I get there ?

• Now we’ll start talking about how to answer these questions.

33 / 48 34 / 48

The pieces we need General control architecture

• Locomotion and Kinematics

How to make the robot move, tradeoff between manoeverability and
ease of control.
• Perception
How to make the robot “see”. Dealing with uncertainty in sensor input
and changing environment. Tradeoff between cost, data and
computation.
• Localization and Mapping
Establish the robot’s position, and an idea of what it’s environment
looks like.
• Planning and navigation
How the robot can find a route to its goal, how it can follow the route.

35 / 48 36 / 48
General control architecture What makes it (particularly) hard

• Changing environment.
• Things change.
• Things get in the way.
• No compact model available.
• How do you represent this all?
• Many sources of uncertainty.
• All information comes from sensors which have errors.
• The process of extracting useful information from sensors has errors

37 / 48 38 / 48

The basic operations Map

• We start with what the robot can “see”.

There are several forms
this might take, but it will
depend on:

What sensors the robot

has

What features can be ex-

tracted.

• (These are not a particularly likely set of features.) • A map then says, for example, how these features sit relative to one
another.

39 / 48 40 / 48
Localization Navigation

• A robot localizes by identifying features and the position in the map

from which it could see them.

• Navigation is then a combination of finding a path through the map. . . .

• Lanser et al (1996)

41 / 48 42 / 48

Navigation How do we put these pieces together?

• A system architecture specifies how these pieces fit together.

• Consider these to be refinements of the “agent with state” from above.
percepts

see state

Environment

next
action

actions

• . . . and avoiding things that get in the way.

• Breaking down next and action into additional pieces.

• Adding in new aspects of state I.

43 / 48 44 / 48
Approach: Classical/Deliberative Approach: Behaviour-based

• Sparse or no modeling
• Complete modeling
• Behavior based
• Function based
• Vertical decomposition
• Horizontal decomposition
• Bottom up

45 / 48 46 / 48

Approach: Hybrid Summary

• A combination of the above.

• Last time we talked about what the main challenges of mobile
• Exactly the best way to combine them nobody knows.
robotics are.
• Typical approach is:
• Let “lower level” pieces be behavior based • This lecture started to describe how we can meet these challenges.
Localization • We covered the main things we need to be able to autonomously
Obstacle avoidance control a robot.
Data collection
• Along the way we looked at how notions of agency — and what this
• Let more “cognitive” pieces be deliberative
Planning
means for autonomy — can help.
Map building

47 / 48 48 / 48