Show HN: XWM β Action-conditioned world models for roboticshttps://github.com/kleyt0n/xwm
Hi HN, I've been working on a new side project to dive deeper into action-conditioned world models for robotics. The goal is to build an open space to train, evaluate, and deploy these models in real environments (primarily for educational and research purposes). In v0, I've put together implementations for several of Yann LeCun's JEPA-based models (I-JEPA, V-JEPA, V-JEPA 2-AC, LeJEPA, LeWM), as well as members of other model families like MuZero and TD-MPC2. I'm sharing it early because I'm open to feedback, code reviews, and collaborations from anyone interested in RL, world models, or roboβ¦