Path planning. RRT & RRT*.
Give two planners the same problem. RRT grows toward random samples; RRT* also revisits the connections it has already made.
Both planners receive the same sequence of samples. Dashed rings include the vehicle’s collision clearance.
What you’re seeing.
My work: multithreaded RRT / RRT* planning in ROS and C++, evaluated in F1TENTH. This browser experiment is a fresh, single-threaded explanation of the underlying search.
The algorithm: sample → nearest node → collision-checked extension. RRT* chooses a cheaper nearby parent and rewires neighbors, propagating cost changes to their descendants. Both runs get the same seeded samples, a 9% goal bias, and the same circular obstacles.
The boundary: this is a 2D geometric planner with a clearance radius. It does not enforce steering curvature or vehicle dynamics. Finite samples can fail to find a route; no finite run is an optimality certificate.
Now make a car follow a path ↗Educational reconstructions of ideas from my work. Demo results are computed here; they are not team test results.