MIT researchers secure multi-robot SLAM with cryptographic protocol CILC
A corrupted robot can reconstruct honest teammates’ imagery from public data—until now.
A team from MIT’s Computer Science and Artificial Intelligence Laboratory (CSAIL) has exposed a critical privacy flaw in multi-agent collaborative SLAM (CSLAM) and proposed a fix. In a new paper, researchers Andrew Fishberg, Yixuan Jia, and Jonathan P. How demonstrate that a corrupted agent within a robot swarm can reconstruct approximations of an honest agent’s camera imagery and trajectory simply by eavesdropping on the global descriptors (GDs) broadcast for inter-agent loop closure detection. While encrypted radios prevent external eavesdropping, they offer no protection against a compromised insider—a threat the team verified concretely.
To close this vulnerability, the team introduces CILC (Cryptographically-secure Inter-agent Loop Closure candidate detection). Instead of transmitting GDs in the clear, CILC uses Secure Multi-Party Computation (SMPC) to compare descriptors privately—only revealing whether a loop closure candidate exists, not the underlying data. The system applies SMPC exclusively to the lightweight step of GD similarity comparison, avoiding the heavy overhead of encrypting the entire CSLAM pipeline. In both simulation and real hardware experiments with visual and LiDAR descriptors, CILC remained real-time and communication-feasible, effectively mitigating information leakage to a compromised swarm member.
- Shows a corrupted swarm agent can reconstruct honest robot’s imagery/trajectory from public GD broadcasts.
- First system to apply SMPC to inter-agent loop closure detection, preserving privacy without full pipeline encryption.
- Validated in real-time on multimodal descriptors (visual + LiDAR) with minimal overhead in simulation and hardware.
Why It Matters
CILC makes multi-robot SLAM privacy-safe against insider threats, enabling secure deployment in sensitive or adversarial environments.