This framework establishes an end-to-end pipeline bridging biomechanics-guided data synthesis and real-time neuromorphic control. By employing a Generative AI, the system expands limited experimental inputs into large-scale gait datasets, which provide the essential variety needed for robust training. These data drive a locomotion controller designed with a neuromorphic architecture that mimics the brain's neural structure, optimizing flexor and extensor oscillators to produce bio-mimetic assistive torques. Once trained, this brain-inspired model operates within a modular hardware architecture, dynamically processing stacked sensor data like IMU and GRF signals. This integration ultimately facilitates adaptive and stable torque control for wearable robots across diverse real-world environments.

This research presents a robust state estimation framework for legged robots, grounded in the theory of the Invariant Extended Kalman Filter (InEKF) on the \( SE_{2+2N}(3) \) Lie group. By explicitly modeling multiple contact points for each leg, the filter accurately captures complex ground interaction dynamics during locomotion. The system's propagation, driven by IMU kinematics, is tightly coupled with a multi-modal correction phase that integrates visual odometry, joint encoders, and pressure sensor arrays. This strategic fusion of proprioceptive and exteroceptive sensors significantly enhances system observability, effectively mitigating drift even in dynamic environments. Consequently, the proposed architecture delivers precise and consistent state estimates essential for stable legged robot control.