Control Flashcards
What are some aspects used to deal with the hardware that control a robot?
- Drive motors
* Read sensors
What are some basic functionalities that control a robot?
- Drive around
- Grasping / object manipulation.
- Processing sensor data.
What are some cognitive functions of robots?
- Task and path planning
- Mapping / world model updates
- Communication / coordination with other entities.
- Learning.
What is the classical control approach?
*Given a current and desired world state, choose the control inputs that produce the goal state.
How many types of control systems do we have for classical control?
*Open loop and closed loop.
How does the open loop control work?
Inputs are computed based on inverted know control law, assumed system states and desired goal states.
*It does not rely in real-world settings.
How does a closed loop control work?
It dynamically adjust control input.
What is the main used of control theory for robotics?
To control actuators.
What are some uses for robot control architectures (RCA)?
- Sense the environment.
- Compute desired state change.
- Drive actuators toward desired state change.
What are the three main types of RCA?
- Hierarchical: Sense - Plan - Act
- Reactive: Sense - Act
- Hybrid : Sense - |Plan| - Act
What is a horizontal RCA?
Data flows through all parts in sequence
What is a vertical RCA?
Data flows through some parts in parallel, typically reactive architectures.
What are deliberative RCA?
Explicit planning and modeling. Use internal state representations.
What are reactive RCA?
Sensor data goes directly to the actuators.
What are behavior-base RCA?
- Typically used in vertical or reactive architectures.
* Basic function implemented as “behaviors”, small modules doing a specific task.
What is the deliberative RCA?
- Classic approach of AI.
- sensors -> Perception -> Modeling -> Planning -> Plan execution -> Motor control -> Effectors -> Sensors
- Perception - cognition - action cycle.
What does modeling know?
- What exists in the world
- Relation between entities
- Properties of entities in the world.
What are examples of logic-based modeling?
- First order logic FOL
- Predicate logics
- Temporal logics
- Spatial logics
- Dynamic logics.
What are examples of non-logic based?
- Procedural representations
- Analogical representations
- Sematic networks
- Action theories
- Frames and scripts.
Common problems of plan execution?
- Incomplete world model
- Effectors fail
- Sensors produce misreadings
- Modification of the world by other agents.
What are the weakness of deliberate RCA?
- Slow turnaround cycle
- Re-planning often necessary due to failures, or unexpected events.
- Reactivity hard to achieve
- Extension of control architecture often difficult.
- Long response time in failure situations
- Limited ability to adapt to environmental changes
- Potential paralysis in highly dynamic environments.
What could be an example of a behavior-based control?
s -> Follow Path -> E
e -> Look for food -> + f
n -> Avoid danger -> f
s -> Wander -> e
o c
r t
^ o
| ————————————– r
How does the behavior-based control work?
- Sensor data goes to all behaviors simultaneuosly
- Each behavior decides on an action
- Behaviors are independent
What is the most simple implementation of a behavior-based control?
Braitenberg vehicle
How is the behavior of a braitenberg vehicle controled?
based on geometrical arrangement of sensors, motors or by a cross connection between sensors.
What are the types of braitenberg vehicles?
- Sensors stimulate same-side motors (Obstacle avoiding)
- Sensors stimulate opposite side motors (Light seeking)
- Sensors go to both motors, some stimulating, some inhibiting. (Depends on weights to achieve behaviors.)
What is a subsumption architecture?
- Layered behavior networks.
- In each layer we define elementary behaviors, aggregate behaviors, continuously select behaviors based on sensor data.
- Limited value of look ahead planning
What are some requirements for behavior-based control?
- Multiple goals
- Robustness
- Extensibility
- Variety of sensors
What are two main reactive architecture styles?
Subsumption architecture
Potential fields
What are the basic vector fields?
- Uniform directional field
- Perpendicular field
- Attraction field
- Repulsion field
- Tangential field
What are the advantages of potential fields?
- Easy to visualize
- Easy to build up software libraries
- Fields can be parameterized
- Combination mechanism is fixed.
What are the disadvantages of potential fields?
- Local minima
* Jerky motion
What is the arbitration via action selection?
The behavior with highest activation level is selected.
What is the voting-based coordination behavior selection?
- A number of committee members receive activation levels.
- Each committee member selects one behavior
- Behavior with most votes wins
What is the behavioral fusion via vector summation?
- Each behavior produces an action vector
* Vectors are summed up
