A Virtual-Physical Interaction Framework for Mobile Robotics Testbeds

Mobile robotics research testbeds play a crucial role in verifying the effectiveness of the robot hardware architecture and control policies, among other aspects. However, physical testbeds are costly and less safe, while virtual testbeds are subject to limitations in simulation algorithms, resulting in lower accuracy. This paper introduces a virtual-physical interaction framework in the context of mobile robotics research testbed that leverages the concept of virtual-physical modeling in digital twins. The framework constructs a bridge that connects the traditional physical hardware platform and virtual simulation testbeds, providing scalable, interactive, high-fidelity simulation and tests on both sides. The proposed framework enables the convenient integration of various applications and experimental scenarios, and it is capable of continuously optimizing the virtual testbed using data from both the virtual and physical testbeds. Robots in the virtual testbed optimized by the proposed framework exhibit dynamics more similar to their physical testbed counterparts, thereby reducing the reality gap to 5%-8% of the maximum value. Furthermore, the proposed framework is generalizable, as the optimized virtual system can perform control tasks that are of a different type but within the same category as the tasks that generated the dataset while maintaining a low reality gap. Extensive experiments demonstrate the performance of the proposed framework.