Feature-Based Robotic Tactile Perception

Roger A. Browse and Susan J. Lederman

Abstract

With the development of array force sensing devices, it is now feasible to consider robotic tactile perception. Experiments with human perception suggest tactile features which can be extracted from force-sensed images. These features form the basis of an object recognition system which exploits constraints on object identity imposed by the knowledge of the location of the force sensor contact.

Introduction

Most existing robotic systems are only capable of operating within a completely understood layout of the workspace. Providing that each situation is a replication of a prototype environment, the robot may perform its actions effectively. Robots of the future will be expected to operate in imprecise and changing environments, and will therefore require sophisticated perceptual abilities.

Considerable effort is now underway in the application of the techniques of 3-dimensional computational vision to robotic situations. There are, however, cases in which vision alone will be insufficient for the development of an understanding of the robotic workspace. Specifically, the robot arm may occlude the imaging camera, and information will be needed about surfaces not facing the camera. Also, detailed knowledge of the characteristics of gripper placement will be critical for effective robotic actions. In response to these needs, tactile sensing devices have been developed, along with computational systems capable of utilizing tactile and kinesthetic information in object recognition.

The approach that we have taken proceeds in three distinct steps: (1) Identify a useful and psychologically valid set of tactile features, (2) Devise computational techniques for the extraction of these features from force sensing devices, (3) Develop techniques which accomplish the recognition of objects and their placement on the basis of these tactile features. An overview of the integration of this tactile perception system with active robotic vision is described elsewhere.