Integrated Ground Reaction Force Sensing and Terrain Classification for Small Legged Robots

Abstract

We present the design and implementation of a miniature tactile sensing array for ground reaction force measurements in small legged robots. Dynamic ground pressure data from the sensors were collected using a small two-legged runner and used to train a support vector machine (SVM) terrain classifier. Results show that tactile sensing data, in combination with information about the motor torque and robot gait, are sufficient to distinguish among hard, slippery, grassy and granular terrain types with >90% accuracy in a single stride. The most useful classifier features include stride frequency, peak motor torque, and peak and average tactile sensor readings.