Hello,

I do not quite get the difference between search and sampling based motion plannings (implemented in the SBPL and OMPL, respectively).

Both use precomputed primitives of the robot to generate a solution. I read the search-based motion planners create a graph from this set of motion primitives and then explores this graph to find an optimum solution, while sample-based motion planners do not create any graph, which leads to faster computations, but less likely to be optimal solutions. How are the motion plans generated by OMPL, then?

I intend to create a car planner to follow a road while avoiding obstacles and keeping inside the lane. The problem here is that I need the planner to operate in real time (check the current planner feasibility or generate a new one at 10Hz, at least). To that end, It would be best to use the OMPL approach and leave SBPL for parking maneuvers, which may require a more exhaustive search, and time is not a problem there.

So, the strategy I want to implement is to use a precomputed set of motion primitives whose radius will be limited by the car's velocity at eath step and check if the chosen plan if the returned plan is feasible by checking an occupancy grid (whose associated cells' coordinated will also be precomputed for each motion primitive to speed up the checking).

Any suggested alternatives are welcome.