Drift-free humanoid state estimation fusing kinematic, inertial and LIDAR sensing

摘要: This paper describes an algorithm for the probabilistic fusion of sensor data from a variety modalities (inertial, kinematic and LIDAR) to produce single consistent position estimate walking humanoid. Of specific interest is our approach continuous LIDAR-based localization which maintains reliable drift-free alignment prior map using Gaussian Particle Filter. module can be bootstrapped by constructing on-the-fly performs robustly in challenging field situations. We also discuss two-tier estimation hierarchy preserves registration this other objects robot's vicinity while contributing direct low-level control Boston Dynamics Atlas robot. Extensive experimental demonstrations illustrate how enable humanoid walk over uneven terrain without stopping (for tens minutes), would otherwise not possible. characterize performance estimator each modality computational requirements.