Lecture 6 - Control Architectures Flashcards
Control Architecture
- definition
- what does it enable?
- A control architecture provides guiding principles and constraints for organizing a robot’s control system (its brain)
- To program a robot in such a way that it will produce the desired result
4 Control Architectures
1. Deliberative Control
- grew out of early AI (playing chess)
-> does not have conflicting outputs
Involves long term planning and therefore thoughtfulness in decision and action
Planning (2)
Process of (1) looking ahead at outcomes of possible actions and (2) searching for the sequence of actions that will reach the goal (shortest path in maze)
Planning
Where does searching happen?
Why not somewhere else?
In robot’s head as a representation
Not in physical world bc robot can search from the goal backwards
4 Control Architectures
- Deliberative Control
- SPA architecture (3)
- Sensing
- Planning
- Acting (executing the plan) -> actuators
4 Control Architectures 1. Deliberative Control Drawbacks - Time scale -> When a problem (2)?
- Many sensors + large internal models = large state space
- > slows down the search - hard if states are partially observable (i.e. location is not completely known)
4 Control Architectures
1. Deliberative Control
Drawbacks
- Space
Memory space
-> robot’s representations must contain all information needed for localizing the robot and planning (distance, images)
4 Control Architectures
1. Deliberative Control
Drawbacks
- Information
- inaccurate representation leads to errors
- > representations must be checked often
4 Control Architectures
1. Deliberative Control
Only useful if..
- The environment..
- The robot knows..
- The robot’s…
- The environment is completely known and does not change during the execution of the plan
- The robot knows what the state of the world is and what the state of the plan is
- The robot’s effectors are accurate enough to execute each step of the plan.
4 Control Architectures
- Reactive Control
- What does it remove?
- What does it use (2)?
- Removes planning component to get fast response
- Uses (1) direct mapping between sensors and effectors. (2) A set of reactive rules couple specific situations to specific actions
4 Control Architectures
- Reactive Control
- When is it used?
- In a dynamic environment
4 Control Architectures 2. Reactive Control Action Selection 1. When needed? 2. Command arbitration 3. Command fusion
- In situations where multiple actions are triggered
- Select one action from multiple candidates
- Combine multiple candidates into a single output action/behavior.
4 Control Architectures 3. Hybrid Control Disadvantages 1. Deliberative Control 2. Reactive Control
- Which need does Hybrid control satisfy?
- Smart but slow
- Fast but inflexible
- Need for effective solutions for different time-scales and representation
Synonym for Hybrid System
and explanation (3)
example
Three-layer system because they consist of three layers
- Reactive layer
- Planning layer
- Middle layer (links 1. & 2.)
example: medicine delivery robot
4 Control Architectures
3. Hybrid Control
What if map changes?
Computing path multiple times?
The planning layer should update the representation
Nah, store the plans!
4 Control Architectures
- Hybrid Control
- Offline planning
- Online planning
- Preplanned plan, for situations that might arise
2. Planning while task is performed
4 Control Architectures
- Hybrid Control
- Why not plan everything?
- Universal plan
- changes in environment/task & too many plans
2. set of all plans, initial states and all goals within state space
4 Control Architectures
- Behavior-Based Control
- What is it?
- Drawback Hybrid System
- Network of internal behaviors which interact in order to produce the desired external behavior (close to reactionism)
- Hybrid has too complex interaction of components
4 Control Architectures
- Behavior-Based Control
- What do behaviors do?
- examples
- What can behaviors do?
- they maintain a goal, executed in parallel
- find object, follow target
- can take inputs from sensors and behaviors AND can send outputs to other behaviors AND can store (decentralized) representations
4 Control Architectures
- Behavior-Based Control
- When should it be used?
- When behaviors in subsystem operate with similar time-scales
- > else: hybrid
4 Control Architectures
- Behavior-Based Control
- Hierarchy of behaviors three layer example
- > what is it called
Layer 0: move around Layer 1: avoid-obstacles Layer 3: find-doors -> subsumption architecture -> bottom up
