Physix Frontier
Physical Intelligence
Products · 3 tracked products
π0.5 Open World Generalization Policy
Upgrade of π0 achieving open-world generalization, enabling mobile manipulators to clean unseen kitchens or bedrooms. Partners utilize this model to solve real-world challenges, demonstrating enhanced adaptability and autonomy in unstructured physical environments.
π0-FAST Autoregressive Action Model
Newly released autoregressive action generation model that improves real-world task success rates and throughput via reinforcement learning. Weights and code are open-sourced, facilitating community development and benchmarking for advanced robotic control strategies.
π0 Generalist Robot Policy Model
First large-scale generalist robot manipulation policy combining multi-task, multi-robot data with novel network architectures. It achieves state-of-the-art dexterity in general manipulation tasks, demonstrating robust capabilities across diverse robotic embodiments and operational scenarios.