Unsupervised Contact Learning for Humanoid Estimation and Control

Nicholas Rotella1, Stefan Schaal2, Ludovic Righetti3

  • 1University of Southern California
  • 2MPI Intelligent Systems & University of Southern California
  • 3Max-Planck Institute for Intelligent Systems

Details

Category

Interactive Session

Sessions

10:30 - 13:00 | Tue 22 May | podH | TuA@H

Humanoids 1

Full Text

Abstract

This work presents a method for contact state estimation using fuzzy clustering to learn contact probability for full, six-dimensional humanoid contacts. The data required for training is solely from proprioceptive sensors - endeffector contact wrench sensors and inertial measurement units (IMUs) - and the method is completely unsupervised. The resulting cluster means are used to efficiently compute the probability of contact in each of the six endeffector degrees of freedom (DoFs) independently. This clustering-based contact probability estimator is validated in a kinematics-based base state estimator in a simulation environment with realistic added sensor noise for locomotion over rough, low-friction terrain on which the robot is subject to foot slip and rotation. The proposed base state estimator which utilizes these six DoF contact probability estimates is shown to perform considerably better than that which determines kinematic contact constraints purely based on measured normal force.

Additional Information

No information added

Video

No videos found

Summary

Walking over rough terrain using contact estimation

  • Investigates methods for estimating continuous, six-dimensional contact probability from data using clustering
  • Uses only proprioceptive sensing (endeffector force/torque and IMU data)
  • Reduces base state estimation error for rough terrain walking tasks
  • Improves stability of closed-loop walking control using base state estimation