Physical Intelligence, one of the most closely followed robotics startups in San Francisco, has introduced a new AI model called π0.7. According to the company, the system allows robots to complete tasks they were never specifically trained to perform. This marks an important step toward a more flexible, general-purpose robot intelligence capable of adapting to unfamiliar situations.
Traditional robot training often relies on teaching one task at a time. A robot learns a specific action, such as folding laundry or making coffee, through repeated examples. Physical Intelligence says π0.7 moves beyond that model by combining skills learned in separate contexts and applying them to new challenges. Researchers describe this as compositional generalization, meaning the robot can remix prior knowledge in practical ways.
One of the most interesting demonstrations involved an air fryer. The robot had almost no direct training data related to the appliance, yet it managed to understand how to interact with it. With verbal guidance from a human, the robot successfully completed the task of cooking a sweet potato. This suggests future robots may be able to enter new environments and improve performance through simple natural-language coaching rather than expensive retraining cycles.
The company also tested π0.7 against earlier task-specific robotics systems and says the new model matched their performance across jobs such as coffee preparation, box assembly, and laundry folding. While researchers acknowledge that robotics still lacks universal benchmarking standards, the results point to meaningful progress.
For business leaders and technology teams, the bigger takeaway is adaptability. If robots can learn workflows through conversation and context instead of custom programming for every task, automation could become faster to deploy and more practical across warehouses, manufacturing, retail, and service industries.
Physical Intelligence has already raised more than $1 billion and is reportedly discussing a new funding round that could value the company near $11 billion. Investor confidence reflects growing belief that robotics may be entering a phase similar to early generative AI, where progress begins accelerating quickly.
We have helped 20+ companies in industries like Finance, Transportation, Health, Tourism, Events, Education, Sports.
