SANDO: Safe Autonomous Trajectory Planning for Dynamic Unknown Environments

A team from MIT led by Kota Kondo has developed SANDO (Safe Autonomous Trajectory Planning for Dynamic Unknown Environments), a breakthrough algorithm that solves the critical trade-off between speed and safety in autonomous navigation. Traditional approaches either use fast soft-constraint planners that can't guarantee safety or slow hard-constraint methods that ensure collision-free paths but are computationally expensive. SANDO bridges this gap through three key innovations: a heat map-based A* global planner that steers paths away from high-risk regions, a spatiotemporal safe flight corridor (STSFC) generator that creates time-layered polytopes, and a variable elimination technique that reduces decision variables in the optimization process.

The system's performance is impressive both in simulation and real-world testing. Ablation studies revealed that the variable elimination technique yields up to 7.4x speedup in optimization time, while the STSFC approach proved critical for maintaining feasibility in dense dynamic environments. Benchmark simulations against state-of-the-art methods showed SANDO consistently achieving the highest success rate with zero constraint violations across all difficulty levels. Most notably, hardware experiments on a UAV demonstrated six safe flights in static environments and ten successful flights among dynamic obstacles using fully onboard planning, perception, and localization systems, confirming robust performance under realistic sensing conditions without ground truth obstacle information.