Chandrayaan-3 uses AI retargeting for safe lunar landing

A new paper published on arXiv (arXiv:2605.29412) reveals the guidance system behind India's Chandrayaan-3 lunar landing. The team, led by Suraj Kumar from ISRO, developed a real-time retargeting policy that uses a convex representation of the controllability boundary. This allows the lander to detect when the original target site becomes infeasible—due to terrain hazards or dynamic errors—and compute a new safe trajectory in milliseconds. The baseline guidance already generated fuel-optimal descent paths, but the high-level retargeting layer adds robustness without compromising efficiency.

The approach is the first data-driven retargeting framework ever used in an operational lunar landing. Pre-flight simulations and actual Chandrayaan-3 flight data validated the method: the system successfully retargeted in real time while maintaining fuel optimality. The paper, accepted at the American Control Conference 2026, demonstrates how machine learning and control theory can merge for safer space missions. By offloading complex feasibility checks to a convex model, it reduces computational overhead, making it suitable for resource-constrained spacecraft. This could set a new standard for future lunar and planetary landers.