Mid-Level Planning and Control for Articulated Locomoting Systems
Technical Report,18 Mar 2011,30 Sep 2016
Carnegie Mellon University Pittsburgh United States
Pagination or Media Count:
We changed directions with the proposed work from midlevel control of articulated systems to design and control of series elastics actuators for snake robots. The ability to adapt to and locomote through unstructured terrain remains a key weakness in the abilities of all snake robots. This problem partly stems from the fact that the gaits that we use to locomote are shape-driven more accurately discrete-curvature driven, and our locomotion is primarily through rolling ground contact. While the shape of the snake can deform significantly to adapt to an environment, our ability to locomote typically depends exploiting structure in the environment through careful choices of the snakes shape. Force control is potentially a good perspective from which to approach the problem of adaptive control and locomotion. One could argue that all of the snake robots locomotion is a result of force interactions, rather than shape interactions. A key development from the MIT Leg Lab to address this problem has been the series elastic actuator. By essentially low-passing the output of the actuator, a series elastic actuator provides more accurate and stable force control and protects the drive components. It also allows practical force sensing using common position-control sensors, such as potentiometers and encoders. We propose implementing a series elastic actuator in the context of our existing snake robots. Specifically, we have built rotational elasticity into to the joints of each snake robot module, and used the rotational displacement of the elastic member to infer the torque exerted by the joint. We have begun developing controllers for this new system.