How is physical AI changing robotics?
A robot that can dance, do backflips, or run a half marathon may grab headlines. But the real breakthroughs could be far less glamorous, like a robot that reliably handles dangerous materials. While traditional industrial robots are great at repeating the same task thousands of times, physical AI aims to give machines more flexibility: the ability to see, interpret, and respond to their surroundings rather than follow a specific script.
So, what happens when you take a robot trained in a virtual world and throw it into a real one? Robots can practice thousands of times in a simulation without breaking a single coffee mug. Put them in a real warehouse, where packages tumble and people walk unexpectedly into view, and things get complicated. Engineers call this the “sim-to-real gap.”4 Better sensors, AI models, simulation, and real-world training are helping bridge the divide.
Despite those challenges, some robots are already earning their keep. At logistics provider DHL, a purpose-built robot named Stretch can unload cartons from shipping containers at rates of up to 700 cases per hour.5 DHL reports that the technology reduces physically demanding work in hot or cold trailers and has announced plans to expand deployments. No backflips required. Just a practical, demanding job, executed by a robot.