Show HN: WorldDiT – a 399M world-action model for simulated robot tasks
Category: ai-ml
Tags: diffusion-models, robotics, reinforcement-learning, world-models, transformer
Score: 7.0/10 (Innovation: 7, Technical: 8, Documentation: 7, Utility: 6)
WorldDiT is a 399M parameter diffusion transformer that unifies world modeling (predicting future RGB patches) and action generation for simulated robot manipulation. It achieves state-of-the-art success on LIBERO benchmarks while being parameter-efficient, and the release includes self-contained inference code, four checkpoints, and evaluation scripts.
Target audience: AI researchers, roboticists, ML engineers
Repository: https://huggingface.co/bageldotcom/worlddit
View on Hacker News