Why Robots Still Struggle With Simple Tasks (And What Might Finally Change That) | Karol Hausman, Co-Founder & CEO of Physical Intelligence

Physical Intelligence aims to solve the problem of general-purpose robotics by creating an AI "brain" capable of generalizing across diverse tasks and environments. Rather than relying on rigid, pre-programmed instructions, this approach leverages large-scale, real-world data and foundation models to enable robots to learn intuitively. A pivotal breakthrough occurred when combining Large Language Models with robotic motion, demonstrated by a robot successfully manipulating objects based on internet-derived knowledge, such as the "Taylor Swift" experiment. By prioritizing real-world data over simulation and integrating reinforcement learning to optimize for task success rather than mere imitation, the company seeks to overcome the reliability bottleneck in robotics. This strategy emphasizes long-term research over short-term commercialization, aiming to build a truly generalist intelligence that can eventually operate across any robot form factor.