Sensor Time Synchronization Simulator

Assumptions and limitations

• Four different things are simulated and reported separately: clock offset (the stamp differs from the true sampling time), measurement synchronisation (pairing frames by stamp), transport latency (arrival − sampling time) and processing delay (added to the same latency here).
• True time is when the world was sampled. Stamp = true time + your offset. Arrival = true time + delay + a uniform random jitter in [0, jitter] (+ the sweep time for the LiDAR). Messages are dropped independently with the probability you set. A fixed seed makes every run repeatable.
• LiDAR: the true time and the stamp refer to the start of the sweep; the message cannot arrive before the sweep ends (1/rate later). Motion distortion inside the sweep (deskewing) is not simulated; only whether IMU data covering the sweep has arrived.
• Pairing: each LiDAR scan takes the camera frame with the nearest stamp inside the slop (ties go to the earlier frame). This mimics the idea of an approximate-time synchronizer, not its exact queue and policy implementation in ROS 2 message_filters.
• Localisation effects are first-order estimates: position error = speed × time error; heading error = turn rate × time error; image shift = f_x·tan(turn rate × time error) with f_x = 450 px. They show the scale of the problem, not what a particular estimator will do.
• Constant speed and turn rate are assumed; real motion changes during the error interval.
• Messages are generated for a fixed 10-second run; the last scan, which ends after the run, is not judged for IMU coverage.