Decision ID: 01
Involved Personnel: Kshitij Kabeer, Harshad Zade, Madhu Korada, and Rolanda Hutson
Date of Revision: 02/09/23
It was decided that we would be going with the Particle Swarm Optimization (PSO) algorithm as our planning algorithm, instead of the more traditional algorithms like A* or RRT* or its variants. The reason for this decision were the problems that we would face if we used more traditional map representations like binary obstacle maps or costmaps, which is what the traditional algorithms use.
The reason why we can’t use these traditional map representations is that a certain obstacle might be an obstacle when the robot approaches it in one direction, and might not be an obstacle when the robot approaches it in another direction. For a more detailed explanation of the problem, please refer to the ILR given in resources.
The PSO algorithm isn’t confined to a particular map representation, since it doesn’t generate trajectories taking into account a discrete map. All it requires is a way to calculate the trajectory cost, which can be done with any map representation. Hence, we decided to go ahead with this algorithm.
Reference documents:
PSO Algorithm Applied to Path Planning
Decision ID: 02
Involved Personnel: Kshitij Kabeer, Harshad Zade, Madhu Korada, and Rolanda Hutson
Date of Revision: 02/22/23
Previously, we had taken a decision to use PSO as our Path planning algorithm. However, after meeting with Professor Maxim Likhachev at CMU (specializes in planning for multiple robotic platforms) we have had to reconsider this decision.
Prof Maxim gave us a lot of insights into how to do planning for our project. The most important insight he gave was that traditional graph based algorithms like A* can be applied to a problem like this if we construct a n-dimensional graph instead of the traditional 2D graph. The number n denotes the number of independent variables of the robot that we want to control. He also outlined the advantages and disadvantages of both the graph based planning approaches and our initial stochastic planning approach.
Now that we have a more nuanced understanding of how graph-based planning can be applied to our problem, we have decided to go ahead with implementing both A* and PSO, and seeing which one performs better, at least in our simulations.
Reference documents: