Mechanical and computational intelligence for agile and robust limbless robotic locomotion in complex aquatic environments

Mechanical intelligence (MI) and computational intelligence (CI) are key components in both organism and robotic locomotion control. MI refers to mechanisms that enhance robustness or simplify locomotion through physical effects, while CI includes perception, information processing, and feedback circuits. We recently showed that MI and CI principles can be applied to overdamped terrestrial limbless systems, modeling C. elegans locomotion [Wang et al, 2023]. However, the importance and roles of MI and CI in inertial and cluttered environments remains unclear, particularly in elongate systems. We thus developed AquaMILR+, a one-meter-long, untethered limbless swimming robot. A bilateral cable-driven system modeling muscle actuation patterns provides controllable anisotropic body compliance. In addition to open-water swimming, MI enables emergent swimming in a heterogeneous aquatic environment (arrays of posts) in open-loop. To improve performance in unpredictable conditions, we implemented real-time compliance tuning using cable tension feedback, boosting robustness and speed. With depth control (via a variable buoyancy mechanism) and gait switching, AquaMILR+ navigated 3D obstacles and performed amphibious transitions. This system can function as a model for studying elongate organism locomotion in cluttered environments (e.g., reefs) and can also inform the design of highly capable all-terrain limbless robots.