Comparing Global Planners for navigation
Dijkstra and A* hit 100% success, but SMAC2D wins on overall performance.
A recent community benchmark compared four global path planners in ROS2 Nav2 — Dijkstra, A*, Theta*, and SMAC2D — across 3,000 randomized navigation tasks repeated over three independent runs. The study used a paired-sample methodology to ensure fairness. Key metrics included success rate, mean computation time, and path quality (length and smoothness).
Results showed a clear trade-off: Dijkstra and A* achieved 100% success rates but produced longer paths (~22 ms average time). Theta* was the fastest (~18.6 ms) and generated the shortest paths, but its success rate dropped to ~86% in cluttered environments. SMAC2D emerged as the best all-rounder, with ~90% success, 19 ms mean time, and smooth, near-optimal paths. Under a real-time 50 ms constraint (20 Hz control loop), all planners mostly complied, but SMAC2D exceeded 99% compliance while Theta* showed the lowest tail latency. The full code and plots are on GitHub, inviting community validation and further refinement.
- Dijkstra and A* achieve 100% success rates but produce longer paths than other planners.
- Theta* is the fastest (18.6 ms) and yields shortest paths, yet its robustness drops to ~86% in cluttered environments.
- SMAC2D offers the best balance: ~90% success, smooth near-optimal paths, and >99% compliance with real-time 50ms constraints.
Why It Matters
Real-world robot navigation needs a planner that is both fast and reliable; SMAC2D emerges as a strong candidate for embedded ROS2 systems.