Comprehensive summary of the Institute for Human and Machine Cognition’s experience with LittleDog

Abstract

We discuss the main issues and challenges with quadrupedal locomotion over rough terrain in the context of the Defense Advanced Research Projects Agency’s Learning Locomotion program. We present our controller for the LittleDog platform, which allows for continuous transition between a static crawl gait and a dynamic trot gait depending on the roughness of the terrain. We provide detailed descriptions for some of our key algorithm components, such as a fast footstep planner for rough terrain, a body pose finder for a given support polygon, and a new type of parameterized gait. We present the results of our algorithm, which proved successful in the program, crossing all 10 terrain boards on the final test at an average speed of 11.2 cm/s. We conclude with a discussion on the applicability of this work for platforms other than LittleDog and in environments other than the Learning Locomotion designed tests.