Learning from unstructured and uncurated data has become the dominant paradigm for generative approaches in language and vision. Such unstructured and unguided behavior data, commonly known as play, is also easier to collect in robotics but much more difficult to learn from due to its inherently multimodal, noisy, and suboptimal nature. In this paper, we study this problem of learning goal-directed skill policies from unstructured play data which is labeled with language in hindsight. Specifically, we leverage advances in diffusion models to learn a multi-task diffusion model to extract robotic skills from play data. Using a conditional denoising diffusion process in the space of states and actions, we can gracefully handle the complexity and multimodality of play data and generate diverse and interesting robot behaviors. To make diffusion models more useful for skill learning, we encourage robotic agents to acquire a vocabulary of skills by introducing discrete bottlenecks into the conditional behavior generation process. In our experiments, we demonstrate the effectiveness of our approach across a wide variety of environments in both simulation and the real world. Results visualizations and videos at https://play-fusion.github.io

Action figures can look a little staid next to video games where your character can walk, talk and fire all manner of weaponry. But there's still something special about the tactile experience of holding a cool character in your hand, which is why we've seen game developers embrace the world of toys with products like Skylanders, Amiibo and LEGO Dimensions. But, while placing a figure on a base can unlock characters or entire worlds, the interaction between game and toy tends to end there. Lightseekers, launching today on Kickstarter, changes that dynamic by making its action figures a living (and almost breathing) part of its games. Lightseekers, in some ways, is almost reminiscent of the film Small Soldiers.