MSSE Project Proposal.
Authors: Max Peysakhov.
Vladlena Galinskaya.
Advisor:William Regli.
Problem:Mechanical Design Evolution.
Scope:Apply Genetic Algorithms to evolution of the mechanical design of robot carrier.
Domain:LEGO Robots.
Deliverables:1) Requirements document
2) Design / architecture document
3) Code
4) Test cases / scripts
Goal:To develop a system whish will accept parameters of the robot carrier and their priorities as input, and will generate optimal or sub-optimal mechanical design of the robot carrier as an output.
Abstract:The idea of this research project is to apply Genetic Algorithm optimization technique to the mechanical designs of the Lego robots. Each design will be represented as Labeled Graph or Labeled Tree. Each design will be evaluated by a set of equations, which we are trying to optimize. Some of the criteria might be:
- Useful Load.
- Speed
- Distance
- Cost
- Consumed Energy
Some other criteria might be discovered to be useful later. Initial population of the robot carrier designs will be generated at random, or by applying complete biased initialization. Given set of the desired ranges for each criteria and their priorities system will evaluate and assign a numeric fitness value to each design. Than candidate new population will be selected by using one of the selection techniques. Single point crossover will be applied if tree design representation is used. Random nodes for two trees will be selected and the sub-tries hanging of these nodes will be swapped. Most likely uniform crossover will be applied if graph representation is chosen. Random mutation will be applied to the population with a certain low probability. Random mutation is equivalent to generating sub-tree or sub-graph at random and swapping it with the part of the design. Translocation mutation, when sub-tries of the same tree switch their places can be used also. This cycle will continue until suitable design is found, or time limit expired.
Project Plan
1 Develop System
1.1 Find and test suitable representation for the mechanical design in the computer
1.2 Find and test suitable set of criteria to evaluate design.
1.2.1 Find and test set of equations describing relationship between criteria.
1.3 Define genetic operators on the design representation.
1.3.1 Initialization.
1.3.2 Selection.
1.3.3 Crossover.
1.3.4 Mutations.
1.4 Tune the parameters for genetic algorithm. (Population size, Mutation and Crossover rates, Selection type, Crossover type, Elitism and other)
2 Develop Documentation
2.1 I will worry about it later.
wrote:
> Prof Mancoridis
> I am going to take my Software engineering graduate project at Summer 99
> and hopefully finish it at Winter 99. Prof W Regli agreed to be my
> project advisor.
> - Could you tell ne what are requirements for this project?
> - What deliverables are?
> - What kind of paper work I and Prof Regli have to submit?
> - How do I register for this project?
> Thank you
> Max Peysakhov
The deliverables should include:
1) requirements document
2) design / architecture document
3) code
4) test cases / scripts
Sign up for indep. study (MSSE Project). You must go to Bob Boyer or
Pat henry to set this up.
Spiros