Recently，he 3-month CVPR 2023 Embodied Intelligence Seminar & Embodied Intelligent Robot Maniskill2 Challenge came to the end. In this Challenge, GXU-LIPE team led by professor Gao Fang of Guangxi Key Laboratory of Intelligent Control, Operation and Maintenance of Electric Power Equipment, College of Electrical Engineering, Guangxi University won three champions in these competition arenas: imitation/reinforcement learning hardware operation , unlimited hardware operation and unlimited software operation.

CXU-LIPE Team is composed of Gao Fang, Li Xuetao, Yu Jun and Shuang Feng. The Team was invited as the champion to submit and attend International Symposium Robotics: Science and System 2023 “Interdisciplinary Exploration of Generalizable Manipulation Policy Learning: Paradigms and Debates”. The GXU-LIPE representative Li Xuetao went to Daegu, South Korea on July 10 to attend the award ceremony, and made a report there. Scholars from Massachusetts Institute of Technology, University of California at San Diego, Berkeley University, Tsinghua University and Google Deep Mind etc. made their reports.

Maniskill series competition are embodied intelligent robot challenges jointly held by University of California at San Diego, University of California at Berkeley and Stanford University. Its first challenge was held in ICLR 2022, the top international conference on machine learning. This year's second Mainskill2 Embodied Robot Competition was held by the CVPR 2023 Embodied AI Workshop, with an award ceremony and academic exchange at the RSS 2023 symposium “Interdisciplinary Exploration of Generalizable Manipulation Policy Learning: Paradigms and Debates”.

The ManiSkill2 Challenge involves large-scale robotic operations aimed at training embodied intelligent robots that can interact with the 3D world like humans.  The challenge could aim at executing different tasks according to diverse objects, such as fill, hang, excavate, pour, pinch, write, stack cube, pick single YCB, pick single EGAD, pick clutter YCB, pick cube, peg insertion side, plug charger, assembling kits, turn faucet and avoid obstacles etc.

Aiming at the overfitting problem of the existing model of Maniskill2 Challenge, the GXU-LIPE team designed a two-stage fine-tuning strategy, which deeply exploited the potential of the model based on point cloud and reinforcement learning network, improved the generalization of the embodied intelligence model, and enabled the robot to autonomously complete simple household chores in virtual environment, which improved the success rate of the robot in completing tasks. The competition also attracted teams from universities at home and abroad, with the second and third place going to teams from Shanghai Jiao Tong University and The Hong Kong Polytechnic University respectively. The award reflects that GXU research team keeps up with the frontier of international disciplines, and constantly achieves international advanced results, which helps the construction of "double first-class" of Guangxi University.