Lecture 4 - Bio-mimetic Robot Control

Question 1

What are the two types of Bio-mimetic robot control?

Answer 1

1 - Central Pattern Generators (CPG)

2 - Behaviour based approach

Question 2

What is the motivation for Bio-mimetic Robot Control?

Answer 2

We want robots to be adaptive.

Current state of the art robots are only able to perform in the exact environment they were designed for.

Additionally, we want them to be robust, i.e adapt to component failure, e.g. a leg on a robot breaks so it learns to crawl.

Question 3

How does a classical robot control its actions?

Answer 3

There is a controller which forces the system, driving the body to a desired location X1. Imagine there is some noise in the system, which causes the system to overshoot, reaching (X1 + delta). Using a feedback loop the error is calculated (delta), the controller is then able to determine and deliver the necessary force to ensure the robot reaches X1.

Examples of such controllers are PID, PD, nonlinear…

Question 4

What are the advantages of classical robot control?

Answer 4

1 - We can control the entire robot from one centralised computer.

2 - We are able to designed optimal control solutions, systems in biology are often non optimal.

Question 5

List the key properties of control in engineering

Answer 5

1- Small number of degrees of freedom

2 - Often simplification of biological systems

3 - Fully actuated

4 - Fully controllable

5 - Model based (rigid body)

6 - Central controller

Question 6

List the key properties of control in nature

Answer 6

• System must adapt to change, e.g failure, growth, you get dench. Lots of ways your body changes and so needs to adapt
• Very complex, very high degree of freedom
• Mostly soft
• Redundancy
• Often model free
• Distributed control

Question 7

What is a CPG?

Answer 7

(Central Pattern Generator) Biological neural networks that produce rhythmic patterned outputs without sensory feedback.

Question 8

Give an example of a CPG in biology?

Answer 8

Consider the example of a cat where the connection between its brain and body is cut. The cat is placed on a treadmill but is still able to walk. A limit cycle is created which enable this rhythmic movement (CPG). Clearly no feedback to the brain.

Question 9

How can a controller change gaits with the use of CPGs?

Answer 9

Consider the CPG as a wave with multiple spikes. If we increase the firing rate, this can induce a change in gait.

Question 10

Give an example of how we could implement an CPG in a robot.

Answer 10

Nonlinear oscillators

Question 11

What does it mean for CPGs to be coupled?

Answer 11

Multiple CPG are influenced by each other, e.g. we want them to be synchronised (our legs during locomotion)

Question 12

What happens if we force the system to leave the limit cycle of a CPG?

Answer 12

It will return to the limit cycle, the system is stable

Question 13

Give examples of the different types of waves which can be induced from CPG’s in biology?

Answer 13

Standing wave (Salamander walking)

Travelling wave ( Salamander swimming)

Question 14

We can increase firing rate to transition from one gait to another. Describe this transition

Answer 14

Smooth transition

Question 15

What is the idea behind behaviour based robotics?

Answer 15

Intelligence is the result of interaction among asynchronous set of behaviour and environment.

Basically it is better to constantly interact with the environment, rather than to using sensors to build a model and execute plans.

Question 16

Describe the four properties of behaviour based robotics.

Answer 16

• Action orientated and reactive
• Bottom up
• Hierarchical
• Adding another layer does not break the system

Question 17

What are the four key aspects of behaviour based robotics?

Answer 17

Embodiment - the body itself plays a crucial role

Situatedness - the physical interaction with the world

No planning required - the world is the best model, so we use it.

Emergent complexity - from layering up simple tasks often more complex behaviour emerges (for free)

Question 18

What is a behaviour?

Answer 18

Reaction to a stimulus.

E.g consider a Roomba robot. It senses a wall, the behaviour is to avoid objects, the response is to reverse.

Question 19

What AI does Fifa use?

Answer 19

A* search

Question 20

Briefly explain an example of a behaviour based robot?

Answer 20

Your lowest level may be to randomly explore. Above this level is obstacle avoidance. You are wandering towards a wall, the obstacle avoidance layer can subsume the wandering layer, correcting the path. You then resume to wandering.

Question 21

Disadvantages to Behaviour based robotics?

Answer 21

Behaviours are intrinsically simple as described, and we cannot rely on emergent behaviours

The behaviour is based on reactions to stimuli, therefore it is questionable if these system actually demonstrate any intelligence.

Question 22

What are some pros of subsumption based architecture? (Hierarchical, behaviour based)

Answer 22

Adaptive
No modelling of the world so electronically ‘cheap’ - computationally and complexity
Emergent behaviour can be better than designed behaviour

Question 23

What are some cons of subsumption based architecture?

Answer 23

Emergent behaviour cannot always be relied upon for specific tasks
Little scope for ‘thinking’ or ‘intelligent reasoning’