Algorithmic control of walking

Abstract
The authors suggest that there are two components required in most control schemes: an algorithmic component which takes a high-level goal and generates joint trajectories, and a dynamics component which takes these joint trajectories and generates the required joint torques. The joint torques are then sent to the mechanism, which (it is hoped) achieves the high-level goal. An approach is presented which simplifies the programming of the algorithmic component for high-degree-of-freedom objects. Instead of supplying a complete set of joint trajectories as a function of time, the algorithm controls other, more intuitive, degrees of freedom. The degrees of freedom are automatically converted to the more conventional joint trajectories. The algorithm can underconstrain the object, in which case constrained optimization is used in converting to joint trajectories. This approach is applied to generate joint trajectories for a walking biped. The walking algorithm, is presented along with the results from testing with the authors' Newton simulation system (ibid., vol.3, p.1806-11).

This publication has 6 references indexed in Scilit: