Energy-Efficient Optimal Control of a Walking Assistive Robot Driven by Human Motion Prediction

Abstract

For fragile individuals with motor disabilities or rehabilitation needs, promoting physical activity while ensuring safe support is crucial. In this context, the growing demand for personalized assistance has heightened interest in robotic devices capable of providing adaptive, physically compliant support during walking. This letter presents an innovative control architecture for a Walking Assistive Robot (I-WANDER), designed to aid mobility, provide stability, and prevent falls in individuals with gait difficulties. 15 healthy participants were asked to walk along diverse paths while using the proposed controller and a classical admittance controller (AC). First, we evaluated the prediction model, which significantly outperformed a Kalman-filter method from the literature (p-value $< $ 0.001). Mean trajectory errors were in the centimeter range, thus demonstrating its suitability for real-time control. Subsequently, we compared the combined LSTM-MPC architecture with the AC and found a significant reduction in energy consumption (p-value $< $ 0.001) and improvements in perceived user effort. Overall, the results demonstrate the potential of integrating data-based trajectory prediction with predictive control to infer user intent, enhance human-robot interaction, and improve motion efficiency in assistive walking scenarios.