Adaptive and Resilient Locomotion in Amphibious Robots: Investigating and Learning from Fishes and Amphibians

Adaptive and resilient locomotion in animals arises from the interaction of neural control, body morphology, sensory feedback, and environmental physics. Understanding how these elements jointly shape locomotor behavior remains a challenge for both neuroscience and robotics, particularly in systems that operate across multiple environments or under perturbations. This thesis investigates the mechanisms that support robust, adaptive, and resilient locomotion in fishes and amphibious robotic systems.

Using neuromechanical simulations and bio-inspired robotic platforms, the thesis examines how spinal pattern generators and sensory feedback generate stable rhythmic motion, how morphology constrains feasible locomotor strategies, and how descending modulation shapes turning and recovery after disruption. To address reward ambiguity in cyclic locomotion, a score-driven inverse reinforcement learning approach is introduced to recover locomotor objectives from neuro-inspired demonstrations and to train feedback policies that generalize across perturbations and environments. Navigation studies in amphibious settings further show that task-level performance depends critically on embodiment and locomotor mode. Together, the results provide principles for designing robotic systems that achieve reliable locomotion through distributed control and embodied interaction.