How a Mac mini powers a staggeringly capable robot without human training
Apple researchers have revealed how they use a Mac mini to control a Unitree humanoid robot to navigate rooms it has never seen before. It's really quite remarkable. The robot can see the table and realize it needs to crouch to get under it - image credit: Apple. Leakers claiming that a future HomeView will have a robot arm didn't envisage anything like this. A team of researchers from Apple has…
Apple researchers have developed a system that allows a Unitree humanoid robot to navigate and interact with unfamiliar living spaces without human intervention. The researchers published a paper detailing their "Humanoid Autonomy, Learned in Simulation, running Onboard" (HALO) system, which utilizes an M4 Pro Mac mini with 64GB RAM to train the robot.
The robot was trained on around 300 hours of simulated data, generated in Nvidia's Isaac Sim, involving 1,000 virtual rooms. Despite never having seen the real-world environments, the robot successfully completed 91% of 360 tasks in six different settings, including following a person, navigating obstacles, and sitting on a chair.
Apple's researchers made the code available on GitHub, but noted that the robot's performance relies on human-written prompts, limiting the speed of task expansion. The system demonstrates the potential for autonomous robotic capabilities without cloud computing or remote control.
Written by urgent.news from AppleInsider's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
