Authors:
Roman Zashchitin
and
Dmitrii Dobriborsci
Affiliation:
Deggendorf Institute of Technology, 93413 Cham, Germany
Keyword(s):
Quadrupedal Robots, Walking Robots, Reinforcement Learning, Gait Generating.
Abstract:
This paper presents an approach for generating various types of gaits for quadrupedal robots using limb contact sequencing. The aim of this research is to explore the capabilities of reinforcement learning in reproducing and optimizing locomotion patterns. The proposed method utilizes the PPO algorithm, which offers improved performance and ease of implementation. By specifying a sequence of limb contacts with the ground, gaits such as Canter, Half bound, Pace, Rotary gallop, and Trot are generated. The analysis includes evaluating the energy efficiency and stability of the generated gaits. The results demonstrate energy-efficient locomotion patterns and the ability to maintain stability. The findings of this study have significant implications for the practical application of legged robots in various domains, including inspection, construction, elderly care, and home security. Overall, this research showcases the potential of reinforcement learning in gait generation and highlights
the importance of energy efficiency and stability in legged robot locomotion.
(More)