December 10, 2010
Robot obstacle avoidance algorithms often have one or more parameter inputs that need to be tuned. Often times the algorithm itself has no guidance for this tuning besides 'guess and check'. This project attempts to solve that problem by both automating the procedure and guiding it using techniques derived from simulated annealing and particle swarm optimization, and is generally applicable to optimizing other functions as well.
- Techniques from Simulated Annealing are used to gradually settle the system from a wide search area to minimums.
- Parallelization ideas from Particle Swarm Optimization are used to help speed up the search process.
The project contains an innovative combination of Simulated Annealing and Particle Swarm Optimization. The primary motivation for this is the function (a simulated test run in this case) execution taking a long period of time without consuming all system resources.
Technology Used Block Diagram
Evaluation of Results
The project was successfully able to derive new values for obstacle avoidance parameters (See the graphs). More testing is needed to see if these values work well in other environments, but the core optimization method worked well and produced parameters that decreased completion time over the old hand-picked parameters.