Learning Energy Efficient Jumping Strategies for Flexible-Legged Systems

Abstract

Legged locomotive systems have many advantages over their wheeled counterparts, such as their ability to navigate rough terrain. There are many techniques to overcome obstacles, one of which is jumping. Still, there are disadvantages to overcome when using legged systems, such as their lack of energy efficiency. To combat this lack of efficiency, flexible links can be used to conserve energy that would otherwise be wasted during locomotion. Furthermore, control methods that improve a jumping system’s ability to jump high and its ability to conserve power can be utilized. In this paper, reinforcement learning (RL) was used to create controllers for a flexible-legged jumping system that maximize jump height while minimizing power usage.