World Model Based Reinforcement Learning

基于世界模型的强化学习

100,000

As Language Model technology (LLM) continues to advance, its integration with robotic systems presents a promising frontier. Beyond its traditional applications, LLM's role extends to enhancing human-robot interactions and collaboration. This project proposes an innovative approach to leverage LLM for optimizing robotic behavior strategy as well as generating realistic training scenario to better transfer the learned knowledge. By employing LLM as reasoning solver, robots can comprehend and perform complex instructions while adapting to different environments. This project also utilizes the concept of world model in term of realist training environment generation for smooth sim-to-real transfer in robotic tasks. The established reinforcement learning framework will become the foundation for next generation intelligent robotic technology.

Large Language Model; Reinforcement Learning; Robotic Learning; Sim-to-Real; World Model