EVENT DOSSIER
The researchers proposed a world model based on Coupled Control and Wireless Joint Embedded Prediction (JEPA) architecture to enhance the robustness of remote robot control. This framework integrates visual observations, radio frequency signal data, and persistent images to simultaneously capture the dynamics of robots and the evolution of wireless channels. Experiments in the Gazebo-ROS-Sionna simulation environment verified the effectiveness of this method. Results showed that compared with traditional PID control, Model-Free Deep Q Networks (DQN), and vision-based Transformer-based prediction methods, it significantly improved communication efficiency, system robustness, and resilience while maintaining navigation performance.
本文提出了一种基于耦合控制与无线联合嵌入预测架构（JEPA）世界模型的鲁棒远程机器人控制框架。该框架利用视觉观测及射频（RF）信号图（Spectrograms）和持久图像（PIs）组合，联合捕捉机器人动力学与无线信道演化。实验在同步的 Gazebo-ROS-Sionna 仿真环境中进行，结果显示该方法相比传统的比例积分微分（PID）、无模型深度 Q 网络（DQN）及基于视觉 Transformer（ViT）的预测方法，显著提升了通信效率、鲁棒性和韧性，同时保持了导航性能。