This is an implementation of the RRTConnect algorithm as described originally here:
James J. Kuffner, and Steven M. LaValle. "RRT-connect: An efficient approach to single-query path planning". IEEE Conference on Robotics and Automation (ICRA), 2000. The implementation contains the following variations from the original algorithm:
-
Greedy sampling of the goal: Every n-th iterations, sample the goal directly. This bias tree growth towards the goal.
-
Greedy 'extend': Instead of extending one 'epsilon' step towards a new sampled configuration, this implementation will make multiple 'epsilon' steps while the new extended node is not in collision and making progress towards the goal.
-
Low dispersion quasi-random sequence (Halton Sequence), for a more uniform and deterministic random sampling of the space, based on the following paper:
Lucas Janson et. al. "Deterministic Sampling-Based Motion Planning: Optimality, Complexity, and Performance", 2016. https://arxiv.org/abs/1505.00023