Our first attempt at learning an artificial visual cortex for embodied agents. Our largest pre-trained model trained on our biggest curated dataset (VC-1) is competitive with or outperforms the best prior task-specific results on all benchmarks that we test on.

Extremely excited to share our new work led by Gunshi Gupta and Karmesh Yadav showing that pretrained diffusion models provide powerful vision-language representations for control tasks that drive efficiency and generalization! All code open-sourced at: github.com/ykarmesh/stabl…

📢Check out our paper ++ Pre-trained Text-to-Image Diffusion Models Are Versatile Representation Learners for Control ++ 📄arxiv.org/abs/2405.05852 The paper was highlighted as a contributed talk at the #GenAI4DM Workshop at #ICLR2024! Join us on Saturday, 11:30am at Lehar 3.