MIT SANDO guarantees collision-free UAV paths in unmapped dynamic environments.

MIT researchers have developed SANDO, an autonomous navigation system that mathematically guarantees collision-free flight for uncrewed aerial vehicles in unmapped, dynamic environments. Published in IEEE Transactions on Robotics, the methodology addresses a critical limitation in existing trajectory planners, which typically fail to provide formal safety assurances when obstacles move unpredictably or when no pre-existing map is available. Lead author Kota Kondo and senior researcher Jonathan P. How, alongside co-authors Jesús Tordesillas, Juan Rached, Lili Sun, and Yixuan Jia, engineered the system to construct time-sensitive safety corridors. Rather than mapping static geometry, SANDO analyzes real-time sensor data to identify dynamic obstacles and calculates their maximum possible velocity. The algorithm then projects predictive spheres around each moving object and builds a protective corridor that accounts for future displacement. A heat map-based planner further optimizes routing by steering the UAV away from high-density obstacle zones, ensuring both efficiency and security. Once the corridor is established, the system employs hard-constrained trajectory optimization to compute the fastest viable path. The architecture enables rapid onboard recalculations, allowing the UAV to adjust its route instantly as the environment shifts. Rigorous mathematical proofs confirm that the planner maintains safety guarantees across all tested edge cases. Experimental validation included simulations where SANDO outperformed leading algorithms in both speed and collision avoidance, followed by twelve physical test flights that successfully navigated real-world dynamic obstacles using only onboard processing. The technology is engineered for high-stakes applications where traditional navigation fails. By operating without prior environmental data, SANDO is suited for search-and-rescue operations in collapsed structures, subterranean mine exploration, wildfire reconnaissance, and autonomous delivery in congested urban zones. Researchers note that future iterations may integrate machine learning to interpret natural language commands and further reduce computational overhead. The development marks a significant advancement in formal safety analysis and spatiotemporal planning for autonomous systems operating in complex, uncharted terrain.