Event

RESPLE2 Multi-Sensor Odometry Framework Released

1 reports · 1 sources

Ziyu Cao and other authors have released RESPLE2, a recursive multi-sensor state estimation framework that represents 6-DoF motion using non-uniform B-splines. It exploits the knot-clamping boundary property to directly predict control points from inertial kinematics, performs prediction and reset within an iterated extended Kalman filter, and in the update step fuses LiDAR, camera, IMU, and leg kinematics at their native timestamps; a separate lightweight global optimization module refines the control points using LiDAR-visual loop constraints. The authors report that, on public and self-collected datasets across handheld, tracked, legged, and air-ground hybrid platforms, RESPLE2 achieves accuracy and robustness comparable to or better than existing multi-sensor odometry systems under the corresponding sensor configurations, especially under aggressive motion, while maintaining real-time operation. The code and experimental datasets have been made publicly available.