Preliminary Work on Compliant Horseshoe Crab Artificial Intelligence Robot (CHAIR)

This paper introduces a compliant robot inspired from horseshoe crab that harnesses the advantages of bendable links to achieve locomotion reminiscent of its natural counterpart. Our robot design features multiple motors for dynamic limb motion and a selectively positioned tail mechanism, which, in future work, will play a pivotal role in swimming. In this initial phase, we focus on establishing fundamental terrestrial locomotion by conducting separate Q-learning experiments for forward movement, left turning, and right turning. Each experiment utilizes gyroscopic feedback to reinforce proper tilt and balance, ensuring stable movement across diverse surface conditions. The success of these individual Q-learning procedures demonstrates how a soft robotic platform can systematically learn essential gait patterns while minimizing disturbances to local ecosystems, owing to the gentle, adaptive interactions afforded by compliant materials. By utilizing bendable links, the robot can conform more closely to uneven terrain, resulting in increased stability and reduced motor strain compared to rigid-body designs. The gyroscope-driven reward structure guides the robot's posture, helping it make timely corrective adjustments for each action. Overall, these results highlight that soft robotics, coupled with reinforcement learning and orientation-based rewards, is a viable approach for developing multi-terrain robots. This work establishes a foundation for more comprehensive action discovery, eventually incorporating additional sensing modalities, continuous tail motions, and further environmental-awareness capabilities to enable fully autonomous amphibious navigation.