A Portable Multi-Sensor Unit and Layered Kalman Filtering for State Estimation on Legged and Wheeled Mobile Robots
12th International Conference on Control, Decision and Information Technologies, CoDIT 2026, Bari, İtalya, 13 - 16 Temmuz 2026, ss.3316-3321, (Tam Metin Bildiri)
- Yayın Türü: Bildiri / Tam Metin Bildiri
- Doi Numarası: 10.1109/codit70676.2026.11631203
- Basıldığı Şehir: Bari
- Basıldığı Ülke: İtalya
- Sayfa Sayıları: ss.3316-3321
- Anahtar Kelimeler: Kalman filter, legged robots, LiDAR odometry, Mobile robots, sensor fusion, state estimation, visual-inertial odometry
- Hacettepe Üniversitesi Adresli: Evet
Özet
We present a portable multi-sensor unit and a layered multi-rate Kalman filtering architecture for real-time state estimation on mobile robots. The hardware integrates a 3D LiDAR, a stereo-inertial camera, an industrial intertial measurement unit (IMU), real-time kinematic GNSS (RTK-GNSS), and on-board computation into a single modular housing that can be mounted, without redesign, on both legged and wheeled platforms. The estimator combines two filters operating at different rates: an error-state Kalman filter (ESKF) that fuses the IMU with leg or wheel kinematics at the IMU rate and an extended Kalman filter (EKF) that anchors the proprioceptive estimate to drift-free exteroceptive sources (LiDAR odometry at ∼1 Hz, stereo visual odometry at ∼15 Hz). The framework is evaluated on a Unitree Go1 quadruped and a Clearpath Husky A200 using OptiTrack ground truth across standard, static-obstacle, and dynamic-obstacle scenarios. Across six logs, the fused estimate tracks the best single-sensor source within a few centimeters and, critically, avoids the catastrophic failures that any single modality exhibits in at least one scenario. We identify covariance miscalibration as the principal cause of the few cases in which a single sensor marginally outperforms the fusion and outline an adaptive-covariance extension as immediate future work.