Walking robots are often constrained by actuator limitations, particularly in space applications where mandatory low-mass designs prevent the use of larger motors. The central objective of this research is the development of gait planning, optimization and control techniques to prevent joint saturation during walking. The effects of compliance on robot behavior and forces/torques are also being studied, particularly its influence on interaction forces at the feet.
Previous research on robotic walking has traditionally included joint torques within gait optimization frameworks either as constraints or with an emphasis on power minimization. In the first case, environment uncertainty and model inaccuracies can still cause torques to be exceeded when executing the motion in real life. In the second case, it is possible to arrive at a torque distribution that reduces power but pushes weak motors closer to saturation (by transferring torque from a big, power-hungry motor to a smaller, power-efficient one).
To address these drawbacks, our approach minimizes the maximum of all joint torques at every step of the motion. This is accomplished by exploiting horizontal motion (XY) and rotation (yaw) of the body. The resulting motion produces reductions of up to 30% in joint torque requirements, and enables uninterrupted walking of robots with limited actuation.
In the optimized gait, a swaying motion naturally emerges. This sway moves the body of the robot away from a leg that is about to step, as intuition would suggest. A similar behavior has been observed in animals such as elephants by researchers who study animal gaits.
Simulation and Hardware
To enable the study of gaits and legged locomotion, we have developed a realistic 3D simulator called GaitView (pictured below). The simulator includes fully interactive visualization, joint-by-joint interaction, and the ability to plan and visualize gaits. It is also capable of parsing and replaying telemetry logs obtained during hardware tests with the real robot. Force and torque analysis functionality is integral to GaitView, and is an essential tool for this work.
Our main hardware testbed has been ATHLETE (also pictured below and at the top of this page), a 6-legged robot developed by the Jet Propulsion Laboratory to provide heavy cargo capabilities during lunar exploration missions. The robot has a mass of 890kg without payload, and stands 2m tall in standard driving configuration. Extensive simulation and hardware experiments have been carried out with GaitView and ATHLETE. Future work might be carried out on other walking robots.
Last modified Fri, 29 Oct, 2010 at 14:37