Physical Intelligence, a startup founded in 2024, builds adaptable AI systems for robots using vision-language-action (VLA) models derived from large language models (LLMs). These models translate general verbal commands into specific actions across different robotic platforms.
Their robots have demonstrated capabilities such as folding clothes, cooking with unfamiliar appliances, and cleaning kitchens. They learn new tasks rapidly, reportedly faster than toddlers.
Despite significant progress and interest from major industry players, key challenges remain. These include collecting varied real-world training data and enabling robots to manage unpredictable human behaviors and environments.
Experts note that while these advances mark important steps toward integrated robots in daily life, profitable and widespread deployment is still a distant prospect.
