Researchers say AI is pushing robots toward more independent work

In short: Robotics researchers say newer AI is helping robots do more on their own at work, but truly flexible helper robots are still far away.

What's going on

Robots in factories have long been good at repeating one motion in a controlled place, like a machine on an assembly line. Now, researchers and companies are trying to build robots that can handle many different tasks in messier, unpredictable places, without constant human control. The International Standards Organization defines robot autonomy as doing intended tasks based on what the robot senses, without human intervention.

Leaders interviewed by Ars Technica said recent AI progress is a big reason for this push. One method is reinforcement learning, which is training by trial and error (like practicing a tennis swing thousands of times). Another is using large “foundation models,” which are AI systems trained on huge amounts of data, so they start with basic knowledge of the world.

Companies are already using more independent robots in limited settings. Boston Dynamics said its Spot robot can do autonomous inspections in hazardous facilities, and its Stretch robot handles packages in warehouses. Agility Robotics has deployed its Digit humanoid robots in warehouses and factories, mostly moving totes and bins, and the company says the robots have logged over 65,000 hours in real operations.

What to watch

Safety is the main blocker to wider use, especially around people. Agility says it has kept robots in separate “work cells” away from workers, and it is working on a new version designed to be safer around humans. Researchers also say training data is still a major gap, because robots need real world practice, not just internet text, and home use is likely still decades away.

Source: Arstechnica