Performance Analysis of RRT* Algorithm

Team Members - Prithviraj Khelkar, Jonas Raedlar, Shiven Sharma

Project for CS599N1 - Robot Brains Course by Prof. Sabrina Neumann

Packages Required

• ompl
• opencv2
• sdl2

Instructions

• The file simpleRRTStar.h contains code for rrt* on stationary scenes, this can be extended to 3 dimensions by adding a new bound in the state space.
• The file armRRTStar.h contains code for rrt* on robots with higher DOF scenes.
• The file PathEngine.h contains code for rrt* on dynamic scenes. It inherits Engine.h
• Extend the class Engine.h to create custom scenes. And extend the class Robot.h to create custom robots.
• globals.h contains important constants, for example DELTA_TIME for the simulation of dynamic scenes.

Goals

Motion planning algorithms play a pivotal role in robotics, with many state-of-the-art algorithms and planners available across multiple platforms and libraries. One of the algorithms is the Rapidly exploring random tree (RRT) and RRT* algorithms, known for its probabilistic completeness and asymptotic optimality. However, RRT* itself introduces computational complexities and limitations. This project presents a comprehensive investigation into the constraints of RRT*. We conduct various experiments on a variety of scenarios in which the algorithm might be used, including 2 and 3-dimensional spaces, as well as stationary and dynamic scenes, using Intel's VTune profiler. Additionally, a state-of-the-art implementation of the Open Motion Planning Library (OMPL) is detailed. The project concludes with insights into the algorithm's performance, laying the foundation for future enhancements and applications.

Data

Stationary Scenes