Complex Adaptive Vacuum Cleaner
From Lazer
Contents |
Overall Description
This is a classifier system or a StarCat system. This system starts out with no initial memory. The system moves around an area and gets positively reinforced for each thing that it collects and negatively reinforced for running into things. This system will have the drive to collect items and to not run into obstacles. The system will have codelets that will try to match its current situation. A situation consists of the inputs and activated codelets. When a codelet matches a situation it will perform it action. This action could be an internal action that trigers other codelet or an external action. When a positive or negative action is performed the codelets that were the cause of this action are reinforced. The system could be reinforced through a bucket brigade. The codelet recognizing string is initialed randomly. Each codelet has a its own strength rating. The strength rating is changed by negative an positive reinforcement. The strength rating is used to give a codelet a better chance of being activated over other codelet. There a cleaning phase that replaces the lowest strength rating codelets with newly created ones. The way that the cleaning phase could be done is by a GA or just randomly creating the new codlets.
The system has to state the immature and mature states. The immature state is what the system starts in. While in the immature state the system lives in a simulated computer environment where it learns it basic skills. After passing some basic steps the system is transfered to the mature state. In the mature state the system lives in a real world environment inside a robot.
Environments
The two environments that the system lives in
Simulation
A computer simulated environment that the system would live in an move around in. Dirt makers - items that produce dirt, but can not be cleaned up. (dog)
- Stationary Dirt makers
- Moving Dirt makers
Robot
An actual robot device that moves around in the real world.
Input to the System
- Vision Camera
- Touching senors
- External touch (bumping into something)
- Internal touch (taste)
- Clock
Training
The system has two stages immature and mature
Immature Stage
The system lives in a simulated computer environment where it learn its basic actions. Some of the basic actions are to avoid obstacles and to keep moving.
Mature Stage
The system is placed into physical robot to perform its same behavior.
Situation
- Inputs
- Activated Codelets
Output to System
- Moving
- Not Moving
Driven Needs of the System
- Collecting Dirt
- Stay Charged
- To Keep moving (maybe)
Reason Why this System is Different
This system has a need to learn, because of its driven need to collect items. This system would have intentionality which is needed for this system to learn. The intentionality of this system would be the desire to collection items.
Other Project with Similarities
- IRobot Vacuum Cleaner
- Classifier Systems robot item picker upper
Pre-Test System
Create the system in the Rule Based Rover environment that was created for the CS 657 class.
References
Equivalent Problems
- Highly moving Rover
- Positive reinforcement for more distance moved
- Negative reinforcement for running into walls
