Path and motion planning

[baggepinnen]

This issue summarizes the available ecosystem when it comes to path and motion planning. We exclude the kind of motion planning that includes dynamics more complicated than a chain of integrators, i.e., we include trajectory generation with acceleration limits, but not with force/torque limits.

The current capabilities of Multibody.jl are documented here https://juliacomputing.github.io/Multibody.jl/dev/#Trajectory-planning

• Path planning: Find a path from one point to another that avoids obstacles. The typical metric is shortest distance, but non-uniform cost over space is sometimes considered.
• Motion planning: Turn a path into a trajectory, i.e., plan the velocity and acceleration profiles with which to follow the path.
• The primary difference between motion planning and path planning is thus that time is not considered in path planning.
• Trajectory optimization / optimal control additionally considers the dynamical model of the system.

Type	Package	Comments
Path	RRTStar.jl	No tests, only 2d
Path	MotionPlanning.jl	No docs. Type unstable code. A couple of different algorithms.
Path	AStarSearch.jl	Only A*, limited docs
Path	Eikonal.jl	Fast marching method (FMM / FSM)
Path	Agents.jl	Only A*
Path	HybridAStar.jl	Only A* (variant)
Path	ProbabilisticRoadMap.jl	Only PRMs
Motion	JSC traj opt	High-order polynomial, 6DOF
Motion	TrajectoryLimiters.jl	Post-processing of heuristic trajectories up to bounded acceleration
Motion	traj5 et al.	Trivial polynomial generation
Motion	JuMP	A lot of motion-planning algorithms can be cast as convex optimization algorithms (often QPs)

https://github.com/alexander-leong/MotionPlanningOptimization.jl

See also

• https://github.com/JuliaComputing/JuliaSimControl.jl/issues/177