Robust and Economical Bipedal Locomotion

Abstract

For bipedal robots to gain widespread use, significant improvements must be made in their energetic economy and robustness against falling. An increase in economy can increase their functional range, while a reduction in the rate of falling can reduce the need for human intervention. This dissertation explores novel concepts that improve these two goals in a fundamental manner. By centering on core ideas instead of direct application, these concepts are aimed at influencing a wide range of current and future legged robots.

The presented work can be broken into five major contributions. The first extends our understanding of the energetic economy of series elastic walking robots. This investigation uses trajectory optimization to find energy-miminizing periodic motions for a realistic model of the walking robot RAMone. The energetically optimal motions for this model are shown to closely resemble human walking at low speeds, and as the speed increases, the motions switch abruptly to those resembling human running. The second contribution explores the energetic economy of the real robot RAMone. Here the model used in the previous investigation is shown to closely match reality. In addition, this investigation demonstrates a concrete example of a trade-off between energetic economy and robustness. The third contribution takes a step towards addressing this trade-off by deriving a robot constraint that guarantees safety against falling. Such a constraint can be used to remove considerations of robustness while conducting future investigations into economical robot motions. The approach is demonstrated using a simple compass-gait style walking model. The fourth contribution extends this safety constraint towards higher-dimensional walking models, using a combination of hybrid zero dynamics and sums-of-squares analysis. This is demonstrated by safely modifying the pitch of a 10 dimensional Rabbit model walking over flat terrain. The final contribution pushes the safety guarantee towards a broader set of walking behaviours, including rough terrain walking.

Throughout this work, a range of models are used to reason about the economy and robustness of walking robots. These model-based methods allow control designers to move away from heuristics and tuning, and towards generalizable and reliable controllers. This is vital for walking robots to push further into the wild.