SCAN-Planner: The Algorithm That Finally Lets Quadruped Robots Navigate Cluttered 3D Spaces Without Collisions

Quadruped robots face challenges navigating narrow passages, cluttered indoor scenes, and large-scale 3D unstructured environments. Existing local planners often approximate the robot with isotropic inflation or rely on planar/elevation maps, leading to conservative motion and poor reasoning about overhanging structures. SCAN-Planner addresses these limitations with a spatial collision-aware framework. It models the robot as a yaw-aware twin-cylinder footprint, enabling precise whole-body collision detection through sparse queries in an inflated 3D occupancy map. This allows the robot to reason about obstacles above and around its elongated body without excessive safety margins.

Key innovations include a projected A* search that generates collision-free guidance on an interpolated ground-following surface, using z-gradient suppression to avoid obstacles horizontally while keeping vertical stability. For large-scale deployment, a robot-centric sliding map with boundary fallback provides high-resolution local collision checking and recovery from local dead ends. Simulation and real-world experiments demonstrate SCAN-Planner produces safe, smooth, and efficient trajectories in dense clutter, 3D unstructured scenes, stair traversal, and long-range navigation tasks. The framework significantly outperforms prior methods in tight spaces and overhanging environments, moving quadrupeds closer to real-world deployment.