Researchers led by Yan Dong have released M-CROSS, a simulated dataset and benchmark for robots that must keep track of their position while moving from land into water and back, according to a paper on arXiv submitted to ICRA 2027. Sensors change abruptly across such transitions, and existing datasets and measures do not capture the resulting failures, the authors write.

The dataset provides synchronised lidar, sonar, stereo camera, inertial, Doppler velocity log and depth data, ground-truth positions, labels for which medium each sensor is in and models of how sensors degrade. New measures track whether a trajectory is completed, accuracy within each medium, errors caused by transitions and how long localisation survives when the robot is partly under water.

Tests with six representative SLAM systems showed that none kept both accuracy and a continuous track through all the water-land transitions.

Source: arXiv Robotics (cs.RO)

Background

Yan Dong

Robotics researcher

Robotics researcher.