Kinematics-based estimation of contact constraints using only proprioception

Abstract

Robots are increasingly being required to perform tasks which involve contacts with the environment. This paper addresses the problem of estimating environmental constraints on the robot's motion. We present a method which estimates such constraints, by computing the null space of a set of velocity vectors which differ from commanded velocities during contacts. We further extend this method to handle unilateral constraints, for example when the robot touches a rigid surface. Unlike previous work, our method is based on kinematics analysis, using only proprioceptive joint encoders, thus there is no need for either expensive force-torque sensors or tactile sensors at the contact points or any use of vision. We first show results of experiments with a simulated robot in a variety of situations, and we analyse the effect of various levels of observation noise on the resulting contact estimates. Finally we evaluate the performance of our method on two sets of experiments using a KUKA LWR IV manipulator, tasked with exploring and estimating the constraints caused by a horizontal surface and an inclined surface.

Vision-Based Framework to Estimate Robot Configuration and Kinematic Constraints

“…This paper extends our recent work on vision-based proprioception and control [12]; and detection and estimation of kinematic constraints [13]. More specifically, in this paper, we show how to combine these two capabilities in a new overarching control framework to address the more general problem of visually controlling an undersensored robot arm, while simultaneously using visual proprioception to detect, explore and model kinematic constraints arising from obstructions encountered by the robot.…”

“…Recently in [13], we presented a method for reliably estimating the kinematic constraints arising from contacts that limit the free motion space of the robot. There we relied on the kinematic observations derived directly from basic proprioception (rotation encoders at joints).…”

Unsupervised Contact Learning for Humanoid Estimation and Control

“…One aspect of contact quality is the determination of contact constraint directions for a given task. Ortenzi et al [9] computed endeffector constraints for a manipulator in contact with a surface using only kinematics, and Nozawa et al [10] recently presented a similar method for estimating environment constraints in humanoid control tasks.…”

Humanoid Whole-Body Movement Optimization from Retargeted Human Motions

“…During the execution of the benchmark, we found out that the sets s 1 − s all did not pick up the box successfully according to Equations (15), (16), (17). Therefore, the initial cost function and improvements for those measures are inexistent.…”

“…Furthermore, demonstrations may be used to learn useful robot control policies. For instance, [14] learn movement constraints from demonstrations, and [15] estimate contact constraints.…”