Getting there: Terrain classification for adaptive legged locomotion over rough terrain

Robots have the potential to save lives by serving as the eyes of rescue workers searching for survivors at a disaster site. With robot assistance, rescuers can access otherwise unreachable spaces while maintaining a safe distance. While today's technology for search-and-rescue is helpful, software and hardware enhancements could significantly improve the utility of these systems.; Current search-and-rescue robots are teleoperated, requiring hours of the operator's full attention. Autonomous navigation through rough terrain can relieve some of this burden. With terrain assessment techniques the robot's motion can be adapted to the terrain. A new, simple approach to terrain classification based on gait bounce is presented. Gait bounce uses vision to measure the tilt of the robot, providing a model of the vehicle/terrain interaction. It is analogous to estimating the roughness of a road based on the bounce of the vehicle.; Distinct terrains generate distinct gait bounce signals. A classifier, which we call a meta-classifier, can distinguish these signals through their spatiotemporal pattern. An extensive study produced 83% accuracy over 700 trials. While this work is conducted with a legged robot, experimental results are presented that show the potential for use on non-legged vehicles.; Intelligent adaptation of motion to the terrain requires the correlation of terrain conditions and motion parameters, relative to a user-defined performance metric. This work outlines the framework for adaptation including an efficiency measure and an adaptive architecture for efficient control.; The main hardware constraints of search-and-rescue are size and agility. Centimeter-scale robots are more easily transported and they can access very small voids. Highly articulated limbs are useful for maneuvering over the rough terrain found inside a felled building and for clearing obstacles. Few robots simultaneously satisfy these constraints of size and agility.; This thesis details a new mechanism TerminatorBot that is ideal for urban search-and-rescue. Its dual-use mechanism capable of both locomotion and manipulation is the key to achieving a small form. Its two 3 degree-of-freedom limbs locomote the body by dragging it forward, utilizing a variety of gaits, and the arms coordinate as fingertips for fine manipulation.