Bio-Inspired Structure and Behavior of Self-Recovery Quadruped Robot with a Limited Number of Functional Legs

Abstract

In this study, the authors focus on the structural design of and recovery methods for a damaged quadruped robot with a limited number of functional legs. Because the pre-designed controller cannot be executed when the robot is damaged, a control strategy to avoid task failures in such a scenario should be developed. Not only the control method but also the shape and structure of the robot itself are significant for the robot to be able to move again after damage. We present a caterpillar-inspired quadruped robot (CIQR) and a self-learning mudskipper inspired crawling (SLMIC) algorithm in this research. The CIQR is realized by imitating the prolegs of caterpillars and by using a numerical optimization technique. A reinforcement learning method called Q-learning is employed to improve the adaptability of locomotion based on the crawling behavior of mudskipper. The results show that the proposed robotic platform and recovery method can improve the moving ability of the damaged quadruped robot with a few active legs in both simulations and experiments. Moreover, we obtained satisfactory results showing that a damaged multi-legged robot with at least one leg could travel properly along the required direction. Furthermore, the presented algorithm can successfully be employed in a damaged quadruped robot with fewer than four legs.

Modeling and Analysis on Energy Consumption of Hydraulic Quadruped Robot for Optimal Trot Motion Control

“…The robot consists of a trunk and four legs. The structure of the legs must be designed considering the fact that an improper design can limit the reachable space of the robot legs when operating in the normal quadruped gaits [11]. Therefore, the leg of SCalf is designed as a 3 DOFs (degree of freedom) structure to make the foot endpoint move in a three-dimension workspace with the hip joint as its origin.…”

Q-Learning of Straightforward Gait Pattern for Humanoid Robot Based on Automatic Training Platform

“…A Q-table is composed of Q-values which are designed and evaluated by the action-value function Q(s, a) for the agent. The Q-values with state s and action a are updated as follows [12,14,16]:…”

“…The ability to improve the desired behavior of the robot is a significant technical challenge. The dynamic motion problems could be solved for unmanned aerial vehicles (UAVs) [11], quadruped robots [12], and even high-dimensional humanoid robots [13] using the Q-learning algorithm [14][15][16]. Most gait patterns are designed for humanoid robots, assuming an ideal situation.…”

Special Feature on Advanced Mobile Robotics

“…The structure design and recovery for a damaged quadruped robot has been tackled by Chattunyakit et al [34]. They showed a caterpillar-inspired quadruped robot whichimitatesthe prolegs of caterpillars, and a mudskipper-inspired crawling algorithm based on reinforcement learning, which improves the adaptation of locomotion.…”