Optimization of Exoskeleton Trajectory Toward Minimizing Human Joint Torques

The reference trajectory, serving as the sole kinematic guidance, is crucial for exoskeleton robot systems. This study introduces a method for generating an optimal trajectory for lower-limb exoskeletons, aiming at reducing human power during walking. Initially, the human joint angles were computed from measured data by a neighborhood field optimization (NFO). Subsequently, inverse dynamic analysis including seven-link dynamic model of human-exoskeleton coupling and corresponding ground reaction forces optimization were constructed, which was surrogated by a back propagation neural network (BPNN) to accelerate successive analyses. The exoskeleton trajectory, generated by perturbing human movement described by Fourier series, was optimized using a NFO algorithm with a revised initial generation strategy and boundary update function to minimize human joint torques. This approach was found to provide more accurate predictions of human trajectory and ground reaction forces compared to traditional methods, achieving a root mean square error (RMSE) within 5 mm and 3 kN respectively, making it suitable for computational applications. The generated trajectory preserves individual walking patterns and anticipates human motion with a mean leading value of 4.6%, effectively reducing joint torque across various gait phases. This research contributes significantly to the analysis of human-exoskeleton interactions and offers valuable insights for designing energy-efficient exoskeletons.