An Intelligent Action Decision Algorithm for Exoskeleton in Edge Networks Environment based on Deep Reinforcement Learning

As wearable devices, ankle exoskeletons can be connected to edge computing networks, this connectivity enables real-time data processing and decision-making at the edge, reducing latency and enhancing responsiveness. The quality of their action decision algorithms significantly influences their effectiveness. Most current algorithms commonly face these issues, including inflexible trajectory tracking and less accurate on-demand assistance, and they did not take into account the noise present in edge networks. To address these challenges, this paper introduces a novel control algorithm named ’SD3-DAE’ (Softmax Deep Double Deterministic Policy Gradients Based On Denoising AutoEncoder). Firstly, the ankle exoskeleton control problem is modeled as a Markovian decision process. Then the SD3 method is introduced to implement the framework of the ankle exoskeleton action decision algorithm for the exoskeleton control problem. Finally, this paper incorporates a Denoising AutoEncoder (DAE) to enhance the SD3 method, improving its robustness and adaptability in the presence of external environmental factors, such as noise. Experimental results demonstrate that our method outperforms other existing model-free algorithms in the realm of ankle exoskeleton control.