An Optical Fiber Proximity Sensor for Haptic Exploration Sean Walker * , Kevin Loewke † , Michael Fischer * , Carl Liu * , and J. Kenneth Salisbury * * Department of Computer Science † Department of Mechanical Engineering Stanford University Stanford, CA 94305 USA {spw, kloewke, mfischer, carlliu}@stanford.edu, jks@robotics.stanford.edu Abstract— This paper presents the design of an optical fiber proximity sensor for haptic exploration with a robotic finger. The sensor uses emitter and receiver optical fiber pairs to measure the intensity of light reflected off surrounding objects in a 2-D workspace. We present the design and construction a 32-point sensor array mounted within a 36 mm diameter finger and describe software techniques to process data acquired by an inexpensive web-cam. We experimentally characterize the sensor performance and demonstrate applications for haptic exploration such as pre-contact velocity reduction and non- contact contour following based on object curvature. I. I NTRODUCTION Much of the dexterous manipulation required in human day-to-day life must be accomplished without visual feed- back. Tasks such as getting keys from a pocket, changing an oil filter, or changing a light bulb all require the use of our exquisite touch capabilities. In robotics, however, the current lack of advanced touch sensing techniques presents a major obstacle to the advance of autonomous and dexterous robotic manipulation. This work is part of an effort to develop next generation dexterous robot hands with new capabilities for sensing, active exploration, and manipulation. In particular, one of the goals of our research is to apply probabilistic techniques [1] to interpret multi-sensory information and model the inherent uncertainty involved with physical manipulation. Accordingly, we have constructed the Probabilistic Manipulation Experiment Table (PMET), as shown in Figure 1, to explore the use of various types of sensors and perceptual methods. The PMET currently con- tains an instrumented two degree-of-freedom robot designed to push objects in the plane with its ‘finger’. The manipulator is controlled by our Probabilistic Robotics Studio (PRS), an interactive design and development studio running Real Time Linux version 3.1 [2] [3]. In this paper we use the PMET and PRS to explore one particular and relatively uncommon perceptual method for robotic manipulation: proximity sensing. Specifically, we present the design and development of an optical fiber proximity sensor (OFPS), as shown in Figure 1. The OFPS has several notable characteristics, including low cost, high sensor density, inherently robust design (as there are no moving parts or mechanical transducers), and high reliabil- ity in controlled environments. One of the most powerful advantages of the OFPS over contact-based tactile sensors is the ability to anticipate an approaching object, which can Fig. 1. Probabilistic Manipulation Experiment Table (PMET). The end- effector, or ‘finger’, is equipped with an optical fiber proximity sensor (OFPS) as well as an embedded accelerometer. make robots cautious and thus more human-friendly [4]. For the robotic manipulator considered in this paper, we use this proximity information for specific tasks such as pre-contact velocity reduction and smooth non-contact contour following based on object curvature. In the following section we discuss previous work related to robot hands, tactile sensors, and optical sensors. We then present the hardware and software development for the OFPS, characterize its performance, and present two experiments which take advantage of its unique capabilities. 2007 IEEE International Conference on Robotics and Automation Roma, Italy, 10-14 April 2007 WeB4.1 1-4244-0602-1/07/$20.00 ©2007 IEEE. 473