The overall goal of this program is to develop a fundamental understanding of the problem-solving activity underlying climbing, then to create new technologies based on this understanding that will enable non-specific multi-limbed robots to free-climb natural, unstructured, vertical terrain. Climbing is regarded by human climbers to be a physical problem-solving activity in a highly unstructured environment. Overall, climbing involves a tight combination of fast but insightful reasoning, goal-directed sensing, and reactive execution. Sophisticated planning is required to handle hard constraints (e.g., equilibrium, torque limits, collision) on the agent's motion, as well as softer ones (e.g., uncertainties, risk level, energy consumption). Precise sensing (e.g., tactile, vision) is used to search and detect potential holds in the unstructured rock face, estimate the location and characteristics of contact points, and anticipate or detect slip. Fine control is needed to maintain balance through careful distribution of contact forces. A solution to the climbing problem requires that these activities be fused into a seamless process.
In collaboration with JPL, the ARL has been researching technologies to enable robotic free climbing (e.g. climbing using only natural features of the terrain, not devices such as pegs in holes). The research in this lab has focused on the motion planning and control problems. We have tested our motion planning algorithms on the JPL's LEMUR robot. We have tested advanced control algorithms in simulation and on the ARL's free flying robots in order to eventually enable their test on LEMUR.
ARL Research on this project has focused in two main areas. The first is that of motion planning, with research by Tim Bretl. He uses probabilistic planning techniques to design a route for a robot to climb. The planned route enables the robot to stay stable at all times while ascending the wall using only friction in the endpoints.
The second area of research, by Teresa Miller, focused on control techniques for a climbing robot. It has been shown in experiments that simple PD control of robotic joint angles is insufficient to deal with the many system uncertainties. For that reason we have investigated types of control which can generate torque trajectories that follow a planned path while minimizing the likelihood of slip.
Last modified Mon, 1 Nov, 2010 at 9:13