Simultaneous Collision Detection and Force Estimation for Dynamic Quadrupedal Locomotion

TL;DR

The paper tackles the problem of simultaneous collision detection and external force estimation for dynamic quadrupedal locomotion using only joint encoder data and robot dynamics. It introduces an interacting multiple-model Kalman filter (IMM-KF) to estimate external wrenches and multiple contact modes (swing, stance, collision) via mode-specific dynamics and pseudo-measurements, independent of gait design. This estimator feeds a reflex motion design, swing-leg admittance control, and a force-feedback MPC to improve collision handling and balancing under disturbances. Experimental results in Gazebo and on hardware demonstrate accurate collision detection and force estimation with real-time performance, improving robustness across different gaits and terrains.

Abstract

In this paper we address the simultaneous collision detection and force estimation problem for quadrupedal locomotion using joint encoder information and the robot dynamics only. We design an interacting multiple-model Kalman filter (IMM-KF) that estimates the external force exerted on the robot and multiple possible contact modes. The method is invariant to any gait pattern design. Our approach leverages pseudo-measurement information of the external forces based on the robot dynamics and encoder information. Based on the estimated contact mode and external force, we design a reflex motion and an admittance controller for the swing leg to avoid collisions by adjusting the leg's reference motion. Additionally, we implement a force-adaptive model predictive controller to enhance balancing. Simulation ablatation studies and experiments show the efficacy of the approach.

Figures (6)

• Figure 1: Force measurement models for different contact modes.
• Figure 2: Gazebo simulation environment for collision tests.
• Figure 3: The likelihoods of different contact modes estimated by the IMM-MBKO. The collision likelihood starts ramping up when a collision happens at around 0.11s.
• Figure 4: Estimated forces in $x$ (upper) and $z$ (lower) directions.
• Figure 5: Desired foot accelerations in the x direction for AC (red) and OSC (blue) before and after the reflex motion (black) is triggered.