This is an old revision of the document!
Notes on robot design - see also lirec notes.
- Michael BradyRobotics is the intelligent connection of perception to action“
- Perception: Sensors tell a robot about the outside world.
- Action: Actuators allow a robot to change the outside world, to act upon it.
- Intelligence: Some sort of controller
State
State is what the robot collects in order to understand the world.
(not to happy about these definitions…)
- Internal state: Directly observable state of the robot. Also it's internal representation of the world - it's internal model.
- External state: What's happening in the real world. Can only be known via a robot's sensors.
- Observable state: known at all times, usually internal - eg. battery state.
- Partially observable state: Through a sensor, along with noise eg: camera image pixels
- Hidden state: What you are actually interested in, in the real world - eg: position of user's head. Can't be directly known, must be indirectly inferred from (maybe multiple) sensors. See sensor fusion.
Controllers
Deliberative vs Reactive
Think hard - then act!
- Capable of learning and prediction
- Finds strategic solution
but:
- Requires lots of computation - eg. brute force search for potential plan
- Needs a world model
- Slow
Don't think, react!
- Fast
- Powerful, biological parallel
but:
- Minimal state
- No memory
- No internal representation of the world
- Unable to plan ahead
- Unable to learn
Or do both!
- Deliberative system on top
- Reactive underneath
- Hybrid or Behavioural system
Links
Artificial Intelligence versus classical Robotics: http://www.ee.pdx.edu/~mperkows/CLASS_479/hardware/014.Robot-Control-Architectures.ppt