Pose estimation for planar contact manipulation with manifold particle filters

Abstract

We investigate the problem of using contact sensors to estimate the pose of an object during planar pushing by a fixed-shape hand. Contact sensors are unique because they inherently discriminate between "contact" and "no-contact" configurations. As a result, the set of object configurations that activates a sensor constitutes a lower-dimensional contact manifold in the configuration space of the object. This causes conventional state estimation methods, such as the particle filter, to perform poorly during periods of contact due to particle starvation.In this paper, we introduce the manifold particle filter as a principled way of solving the state estimation problem when the state moves between multiple manifolds of different dimensionality. The manifold particle filter avoids particle starvation during contact by adaptively sampling particles that reside on the contact manifold from the dual proposal distribution. We describe three techniques-one analytical, and two sample-based-of sampling from the dual proposal distribution and compare their relative strengths and weaknesses. We present simulation results that show that all three techniques outperform the conventional particle filter in both speed and accuracy. Additionally, we implement the manifold particle filter on a real robot and show that it successfully tracks the pose of a pushed object using commercially available tactile sensors.

Shape and pose recovery from planar pushing

“…More recently, Koval et al [15] introduced the manifold particle filter, imposing a hard constraint on the binary contact/no-contact condition when tracking the pose of an object.…”

Tactile manipulation with biomimetic active touch

“…The situation bears analogy with utilizing a Kalman filter to combine information for accurate control, which is linked with recent tactile manipulation methods based on classical control theory [14] and particle filters [17]. Our expectation is that understanding the relation between these approaches and the biomimetic approach proposed here will help in solving the overall problem of attaining robust and general tactile manipulation in complex and uncertain environments…”

“…Tactile sensors have been used for object recognition [12], improving grasp stability with force control [13] and object exploration/manipulation through edge or surface following [14]. Tactile servoing [15] has been applied to object manipulation on an industrial robot arm [16] and particle filter methods for controlling how to push objects using tactile feedback [17]. Bayesian methods have been proposed for in-hand manipulation [18], [19]; here we examine tactile manipulation from the perspective of biomimetic active perception.…”

Tracking objects with point clouds from vision and touch

“…A parallel body of work employs particle filters (PFs) to tackle exactly the ambiguity and nonlinearity often inherent in contact state estimation. Koval, et al take advantage of the manifold structure of the state space of contact to greatly reduce the critical particle starvation issue facing PFs during contact events by resampling directly from the contact manifold [11]. Zhang and Trinkle tackle the same problem by using a constraint-based physical model to enforce that particle updates stay physically feasible with respect to nonpenetration and contact forces [25].…”

“…Schmidt, et al support a binary contact detection signal as an input, and estimates contact locations that explain the binary contact detections [19]. The contact manifold [11] [10] and contactmode switching [14] approaches are natural when used with a discriminative sensor, but do not extend as naturally to dense geometric contact information.…”