Unnecessary Movements and Timeouts: LazyPRMstar vs. TRRT Planner in ROS2 and Moveit2
We are currently using ROS2, Moveit2 Humble and an ur10 to grab items from a table.
For path planning we tried to use the LazyPRMstar planner, resulting in completely unnecessary movements and lot of failed plannings because of Time Outs (in the end we set the planning time to 15s to reduce the timeouts - with mixed results). As an alternative we tried the TRRT Planner, resulting in very efficient planning (0.2s) and paths and sometimes some minimal unnecessary movement.
To objectively justify our our decision against LazyPRMstar I tried to use the moveit_ros_benchmark package to compare the two planners. However, this comparison turned out to be in favor of LazyPRMstar.
The only notable difference I'm aware of between our implementation and the benchmark package is that we use the Move Group Interface, whereas the benchmarking uses the C++ Interface.
I would appreciate any insights or speculation as to why LazyPRMstar underperforms outside of the benchmark scenario.
Thanks in advance.
