The article introduces RoboGen, a generative robotic agent that learns a variety of robotic skills through generative simulation. It uses advancements in foundation and generative models to produce a range of tasks, scenes, and training supervisions. The agent proposes tasks and skills, generates corresponding simulation environments, and then learns policies to acquire the proposed skill. The system can be queried repeatedly, generating a continuous stream of skill demonstrations for diverse tasks and environments. The goal is to transfer extensive knowledge embedded in large-scale models to the field of robotics.
Publication date: 2 Nov 2023
Project Page: https://robogen-ai.github.io/
Paper: https://arxiv.org/pdf/2311.01455
