Simultaneous State Estimation and Contact Detection for Legged Robots by Multiple-Model Kalman Filtering

Marcel Menner,Karl Berntorp

2024-04-04

Abstract:This paper proposes an algorithm for combined contact detection and state estimation for legged robots. The proposed algorithm models the robot's movement as a switched system, in which different modes relate to different feet being in contact with the ground. The key element in the proposed algorithm is an interacting multiple-model Kalman filter, which identifies the currently-active mode defining contacts, while estimating the state. The rationale for the proposed estimation framework is that contacts (and contact forces) impact the robot's state and vice versa. This paper presents validation studies with a quadruped using (i) the high-fidelity simulator Gazebo for a comparison with ground truth values and a baseline estimator, and (ii) hardware experiments with the Unitree A1 robot. The simulation study shows that the proposed algorithm outperforms the baseline estimator, which does not simultaneous detect contacts. The hardware experiments showcase the applicability of the proposed algorithm and highlights the ability to detect contacts.

Robotics,Systems and Control

What problem does this paper attempt to address?

The problem this paper attempts to address is the development of a simultaneous state estimation and contact detection algorithm for legged robots. Specifically, the authors propose a method based on the Interacting Multiple-Model Kalman Filter (IMM-KF) to identify different foot-ground contact situations by modeling the robot's motion as a switching system, and on this basis, estimate the robot's state. This approach aims to overcome the cost increase and fault sensitivity issues associated with existing methods that rely on additional sensors (such as force sensors, cameras, etc.), using only Inertial Measurement Unit (IMU) sensor data (Euler angles, rotation, linear acceleration) and motor measurement data (joint angles, joint acceleration, motor torque). The effectiveness of the proposed algorithm is validated through high-fidelity simulator Gazebo and hardware experiments, demonstrating its potential application on uneven terrain. The study shows that compared to baseline estimators, this algorithm can perform state estimation and contact detection more accurately.