Jiajun Cao, Kai Tang, Bohua Zhang, Zezhong Qian, Ying Li, Xiaozhu Ju, Xiaobao Wei, Jiaming Liu, Yang Ding, Sixiang Chen, Yiding Sun, Siyuan Qian, Tinghao Wang, Jixian Wu, Yichen Guo, Yi Zhang, Shilong Zou, Yong Dai, Jian Tang, Shanghang Zhang
The LAW framework unifies language, action, and world modeling for agentic physical intelligence, achieving 94.21% average success across 50 RoboTwin tasks. LAW outperforms existing models, including Embodied-R1.5 and GPT-5.5 High, despite having only 2B parameters. This framework enables more effective long-horizon manipulation tasks.
The LAW framework has significant practical implications for robotics and artificial intelligence, enabling more efficient and effective completion of complex tasks. This technology can be applied to various real-world scenarios, such as manufacturing, healthcare, and service industries. By improving the success rate of long-horizon tasks, LAW can enhance productivity and reduce errors.
Thank you to arXiv for use of its open access interoperability. arxiv.org. aibrief.ai is not affiliated with or endorsed by arXiv.