luokai0 Path Planning Expert
Before Starting
- Global or local planning?
- Known or unknown environment?
- Holonomic or non-holonomic robot?
Core Expertise Areas
Configuration Space
C-space: space of all robot configurations, obstacle-free region is C-free. C-obstacle: set of configurations causing collision with obstacles. Dimensionality: one dimension per DOF, high-dimensional for complex robots. Completeness: planner finds path if one exists, or reports none if not.
Sampling-Based Planners
PRM: probabilistic roadmap, sample configs, connect nearby ones, query multiple times. RRT: rapidly-exploring random tree, single query, extends tree toward random samples. RRT-star: asymptotically optimal, rewires tree to minimize cost. Bidirectional RRT: grow trees from start and goal, connect when they meet. Informed RRT-star: focus sampling in ellipsoidal subset once solution found.
Graph Search
Dijkstra: optimal path in weighted graph, explores all nodes within cost bound. A-star: heuristic-guided search, optimal with admissible heuristic. D-star: dynamic A-star, replans efficiently as new obstacles discovered. Lattice planners: regular grid in C-space, supports non-holonomic constraints.
Trajectory Optimization
CHOMP: covariant Hamiltonian optimization for motion planning, gradient descent. STOMP: stochastic trajectory optimization, samples noisy trajectories. TrajOpt: sequential convex optimization, handles collision avoidance as constraint. Time-optimal trajectory: minimize time subject to velocity and acceleration limits.
Best Practices
- Choose planner complexity to match problem dimensionality
- Tune sampling distribution to improve planner efficiency
- Post-process planned paths to remove unnecessary detours
- Validate planned trajectories against dynamic feasibility constraints
Common Pitfalls
| Pitfall | Fix |
|---|---|
| Planning in task space for high-DOF robots | Plan in C-space to handle all constraints |
| Ignoring dynamic feasibility | Check velocity and acceleration limits on planned path |
| Inadequate collision checking | Use conservative bounding volumes for safety |
| RRT tree biased to initial region | Ensure uniform random sampling of C-space |
Related Skills
- robot-kinematics-expert
- control-theory-expert
- computer-vision-robotics-expert