Jun Xiong

Jun Xiong

Don't be blue.

Publications

Patent

A movable IoT security monitoring device

Published in CNIPA, 2019

The utility model discloses a movable Internet of Things security monitoring device, comprising a monitor, a base device, the monitor being installed on the lower side of the base device, the base device comprising a base body, the interior of the base body being provided with a base groove, and the interior of the base groove being penetrated by a steel wire. The interior of the base groove is provided with two rotating shafts, and the two ends of the two rotating shafts are embedded in the interior of the base body. Both rotating shafts are equipped with rollers, and the rollers are provided with roller grooves. The steel wire passes through the interior of the roller grooves. The interior of the base body is provided with a cavity, and the interior of the cavity is provided with gears. The first pulley is installed on the rotating shaft located on the left side, and the second pulley is installed on the rotating shaft located on the right side. A belt is connected between the second pulleys, and a motor is installed inside the cavity, The output shaft end of the motor is equipped with a worm, which meshes with the gear. The IoT security monitoring device is movable and has a wide monitoring range.

Recommended citation: 1. Jun Xiong. A movable IoT security monitoring device[P].F16M11/42（2006.01）:CN209725749U,2019-03-28.
Download Paper

Conference Papers

RealMirror: A Comprehensive, Open-Source Vision-Language-Action Platform for Embodied AI

Published in ICRA, 2025

We propose RealMirror, a comprehensive, open-source embodied AI VLA platform. RealMirror builds an efficient, low-cost data collection, model training, and inference system that enables end-to-end VLA research without requiring a real robot. To facilitate model evolution and fair comparison, we also introduce a dedicated VLA benchmark for humanoid robots, featuring multiple scenarios, extensive trajectories, and various VLA models. Furthermore, by integrating generative models and 3D Gaussian Splatting to reconstruct realistic environments and robot models, we successfully demonstrate zero-shot Sim2Real transfer, where models trained exclusively on simulation data can perform tasks on a real robot seamlessly, without any fine-tuning.

Download Paper