An Imitative Approach to Adaptive Gait Generation for Lower Limb Exoskeletons through Kernelized Movement Primitives

The development of lower limb exoskeletons (LLE) has gained significant attention due to their potential applications in assistive technology and human augmentation. Adaptive gait generation is a critical component for enabling LLEs to operate safely and naturally in real-world environments. This thesis presents an imitative approach to adaptive gait generation based on Kernelized Movement Primitives (KMP). By leveraging a limited dataset of gait demonstrations from level-ground walking, the proposed framework learns joint-space and task-space trajectories that closely mimic natural walking patterns. A depth camera mounted on the exoskeleton provides real-time environmental data, enabling dynamic adjustments to the gait cycle for scenarios such as obstacle avoidance, stair negotiation, slope traversal, and variable step lengths. To ensure that the generated trajectories remain both physiologically plausible and safe, the framework integrates a linearly constrained local-KMP formulation alongside a Max Knee Angle Reduction (MKAR) algorithm. These enhancements enforce critical biomechanical constraints while preserving the natural dynamics of human gait. Extensive simulations and experimental validations demonstrate that the imitative approach not only replicates the desired movement patterns with high accuracy but also offers improved adaptability and robustness compared to traditional analytical methods. This work advances the field of assistive robotics by delivering a data-driven and flexible solution for gait generation. Ultimately, the proposed methodology contributes to more natural and reliable human-exoskeleton interactions, enabling exoskeletons to be safely and independently used in everyday environments.