How do you design a robust multi-sensor SLAM system for autonomous vehicles?

Multi-sensor SLAM design: 1) Sensor selection - lidar for 3D structure, camera for semantic information, IMU for high-rate odometry, GPS for global reference, wheel encoders for scale; 2) Calibration - precise extrinsic calibration between all sensors, temporal synchronization handling different update rates; 3) Frontend processing - feature extraction and matching per modality, tight coupling in factor graph or loose coupling with separate odometry sources; 4) Backend optimization - graph-based or filter-based state estimation, loop closure detection and correction, multi-session mapping; 5) Failure handling - sensor dropout detection, graceful degradation maintaining localization with reduced sensors. Challenges include computational efficiency, handling dynamic objects, and long-term map consistency. Validation requires diverse environments and conditions.