Decoupled state estimation for humanoids using full-body dynamics

Abstract

We propose a framework to use full-body dynamics for humanoid state estimation. The main idea is to decouple the full body state vector into several independent state vectors. Some decoupled state vectors can be estimated very efficiently with a steady state Kalman Filter. In a steady state Kalman Filter, state covariance is computed only once during initialization. Furthermore, decoupling speeds up numerical linearization of the dynamic model. We demonstrate that these state estimators are capable of handling walking on flat ground and on rough terrain.

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

“…The influence of an erroneous velocity is transient and quickly corrected by subsequent observations. Using both measurement types together would be comparable to the approach in [2], but we avoid doing so as it raises the possibility of creating inconsistencies, particularly when combined with position measurements derived from the LIDAR module (presented in the following section).…”

“…Errors in link center-of-mass modeling and the presence of unpredictable forces (e.g., from the robot's support/power tether or external contacts) are accounted for by appending an additional process model for each class of disturbance. mfallon,antone,nickroy,teller@csail.mit.edu In [2], the authors extend this approach and apply it to the Atlas robot (which we are also using in our work). They discuss the computational challenges of formulating a single extended Kalman filter (EKF) for a humanoid with many degrees of freedom, and propose instead to estimate the pelvis position and joint dynamics in separate filters.…”

“…This cloud is then converted into a probabilistic occupancy grid (OctoMap [15]) against which efficient localization comparisons are later performed. While the MAV experiments presented in [12] required offline mapping with a separate sensor, our legged humanoid and actuated LIDAR with 30 m range permit the map to be constructed immediately prior to operation and utilized during the entire task 2 . This makes our approach practical both for highly variable laboratory experiments and for field trials in which the environment is initially unknown.…”

“…The estimator does not require elaborate dynamics models; like [4], we couple foot placements and leg kinematics with inertial predictions in an EKF framework, and like [2], we consider pelvis pose separately from joint states. Our primary novel contributions are (1) incorporation of exteroceptive sensing to achieve reliable drift-free alignment to a prior map while walking using a Gaussian Particle Filter (GPF) to apply position corrections derived from each LIDAR scan; and (2) extensive experimental validation of the algorithm on a real humanoid robot.…”

“…We exclude both the robot's contact foot position and the joint states from this state vector and filter them separately (as do others including [2]). The pelvis of the Atlas humanoid contains the robot's primary Inertial Measurement Unit (IMU), located 9 cm behind the pelvis link position and rotated by 45…”