Autonomous Robots

1
Autonomous Robots

• Robot Control Architectures

2
Robot Control Architectures

• Robot control architectures are needed to allow
  robots to coordinate their behavior
• How can a robot generate intelligent behavior
  to address complex tasks ?
• Deliberative Control Systems Intelligent
  behavior can be generated using complex programs
  which plan how to achieve an outcome.
• Cyberentics/ Reactive Control Intelligent
  behavior can be generated by a combination of
  simple closed-loop controllers interacting with
  the world.

3
Robot Control Architectures

• Spectrum of Control Approaches

• Good for situations with full knowledge about the
  environment
• Assembly / Factory applications

• Good to address unforeseen situations and where
  fast response is needed
• Noisy environments / real world

4
Deliberative Robot Control Architectures (recap)

• In a deliberative control architecture the robot
  first plans a solution for the task by reasoning
  about the outcome of its actions and then
  executes it
• Control process goes through a sequence of
  sencing, model update, and planning steps

5
Deliberative Control Architectures

• Advantages
• Reasons about contingencies
• Computes solutions to the given task
• Goal-directed strategies
• Problems
• Solutions tend to be fragile in the presence of
  uncertainty
• Requires frequent replanning
• Reacts relatively slowly to changes and
  unexpected occurrences

6
Reactive / Behavior-Based Control

• Construct robot control from a set of concurrent
  behaviors
• Behaviors represent direct stimulus-response (i.e
  senor-actuator) mappings
• Complex task performance as a result of the
  combination of all behaviors as they interact
  with the environment
• Each behavior only requires a partial view/model
  of the world

7
Behavior-BasedRobot Control Architectures

• In a behavior-based control architecture the
  robots actions are determined by a set of
  parallel, reactive behaviors which map sensory
  input and state to actions.

Behavior 1
Behavior 2
Behavior 3
Composition
Actuators
Sensors
Behavior 4
?
Behavior n
8
Behavior-BasedRobot Control Architectures

• Advantages
• Robot can react fast to changes
• System does not depend on complete knowledge of
  the environment
• Behaviors are easier to design than complete task
  strategy
• Problems
• Emergent behavior (resulting from combining
  initial behaviors) can make it difficult to
  predict exact behavior
• Difficult to assure that the overall task is
  achieved

9
Behavior-BasedRobot Control Architectures

• Design Issues
• How can behaviors be represented ?
• How can the right set of behaviors be designed ?
• What is a good coordination mechanism ?

10
Behaviors

• Navigation behaviors can often be represented by
  potential fields

• Obstacle avoidance

• Goal reaching

11
Navigation Behaviors (cont)

• Obstacle circumnavigation

• Random walk

12
Navigation Behaviors (cont)

• Overall behavior from combining navigational
  behaviors

13
Behaviors

• Each behavior can have its own partial
  model/representation of the world
• E.g. Obstacle avoidance
• Model/Representation Sonar range ahead of robot
• Relative obstacle location
• E.g. Move to goal
• Model/Representation heading and distance
  towards goal
• Limited model requirements make behaviors more
  robust to noise and improve their response time

14
Behavior Construction

• How can the right set of behaviors be constructed
  ?
• Often Design Evaluate Modify

Construct minimal system
Run robot
Evaluate performance
Add special new behavior to address problem
Correct behavior ?
No
Yes
Done
15
Behavior Representations

• Different means of representing behavior
  combinations are used
• Stimulus-Response Diagrams
• Indicates stimulus-response mapping of the
  behaviors

16
Behavior Representations

• Finite State Acceptors
• Indicates switching between different behaviors

17
Behavior Example

• The behavior of, e.g., a foraging robot can be
  divided into elemental behaviors that, when
  combined correctly, solve the task
• Foraging sub-tasks
• Wander Moves through the environment
• Avoid Avoid running into obstacles
• Move to food Move towards a food item
• Pick up Pick up food
• Home Move to the home location

18
Coordination Mechanisms

• The coordination mechanism determines what signal
  to send to the actuators given the individual
  (potentially contradictory) commands from all of
  the behaviors
• Basic coordination principles
• Competitive (winning behavior drives robot)
• Priority-based
• Action selection
• Finite State Sequencers
• Cooperative (mixture of behaviors drives robot)
• Weighted sum of responses

19
Subsumption

• Subsumption architecture is one of the earliest
  behavior-based architectures
• Behaviors are arranged in a strict priority order
  where higher priority behaviors subsume lower
  priority ones as long as they are not inhibited.
• A behavior is active (i.e. produce output) if
  input conditions of the behavior are met and it
  is not inhibited by another behavior
• Active behaviors can inhibit or subsume other
  behaviors

20
Subsumption

• Lower layer behaviors can inhibit higher levels
• Higher layer behaviors can subsume other
  behaviors
• Foraging example

Pure priority-based arbitration
21
Subsumption

• Higher layer actions subsume lower layer
  behaviors (highest layer active behavior drives
  the robot)
• Behaviors have to have precisely designed
  conditions under which they are active
• Wander always active
• Move to food active if the robot sees the food
• Pick up active if food is within reach
• Home active if the robot has food
• Avoid active when an obstacle is in front of the
  robot

22
Subsumption

• Priority-based coordination is easy to design but
  very sensitive to the right order of priorities
• Inverting the priority between Move to food and
  Avoid can significantly change behavior
• If Avoid is higher, the system might not be able
  to reach the food if it is close to an obstacle
• If Move to food is higher, the system might hit
  an obstacle if it is close to food

23
Subsumption

• To address problems in task performance,
  additional, higher priority behaviors can be
  added that have specialized activation conditions
• Design-Evaluate-Modify cycle can be performed
  relatively easily
• New behavior to address problem is added on top
  of old ones
• Activation condition is exactly the description
  of the conditions under which current behavior
  set fails
• Behavior is a specific control law to solve this
  problem

24
Subsumption

• Advantages
• Very simple coordination mechanism
• Behaviors can be very specific
• All behaviors can run in parallel
• New layers can be added easily
• Overall behavior can be predicted relatively
  easily
• Problems
• Determining the right prioritization between
  behaviors can be difficult
• Strict priorities can require very large numbers
  of specialized behaviors (behavior proliferation)

25
Action Selection

• Action selection replaces the fixed priorities of
  the behaviors with activation numbers that each
  behavior calculates
• Activation number is a measure of the behavior as
  to its current urgency to drive the robot
• Coordination mechanism selects the behavior with
  the highest activation number to actuate the
  robot
• Behaviors can be always active (instead they can
  lower their activation output

26
Action Selection

• Foraging example
• Wander Activation C1, C1ltC4ltC3
• Move to food Activation C2/(max(distance to
  food, reach)
• Pick up Activation C2/reach ? -f object is
  within reach, otherwise
• Avoid Activation C3/(distance to obstacle)2

27
Action Selection

• Dynamic priorities (activation numbers) allow
  behaviors to change their priority order
• E.g. Avoid is higher priority than Move to food
  if the distance to the obstacle is much closer
  than the distance to the food. Otherwise Move to
  food might be higher priority
• Fewer behaviors can more flexibly address tasks
• In subsumption, the only way to move to the food
  is if avoidance is not active (i.e. food can only
  be reached if it is outside the activation
  condition for Avoid). To solve this would require
  another approach behavior that is active if the
  food is in front of the obstacle

28
Action Selection

• Advantages
• Relatively simple coordination mechanism
• Dynamic calculation of priorities (allowing for
  different behaviors to be important based on
  sensory input
• All behaviors always run in parallel
• Relatively easy to predict the behavior of the
  system
• Problems
• Activation functions have to be designed
  carefully