Imagine a robot that could do your laundry, make your bed, cook dinner or stock a grocery shelf. Humans have long taught robots individual tasks, but instructing them to handle more complex, changeable jobs has been elusive despite billions invested in robotics.
A team at École Polytechnique Fédérale de Lausanne in Switzerland, led in part by robotics scientist Sthithpragya Gupta, reports progress. They published a paper in Science Robotics describing a machine-learning method that uses kinematic intelligence — a robot’s built-in awareness of how its body can move — to learn from human demonstrations. In videos of the work, single-arm robots watch a person toss a ball into a container, then copy the behavior while adjusting for their own positions and nonhuman bodies. The learned skills can be transferred to other robots.
Gupta says this could enable robots to handle subtler, variable tasks — like making coffee to someone’s taste — rather than only repeating narrowly programmed actions. He and others note a long-standing problem: robots can execute a trained action reliably, but small changes in context (an opponent moving in tennis, lighting shifts) can break performance. Teaching adaptation from humans to robots has been difficult; the new approach addresses that gap.
Experts see the work as a breakthrough. Robert Platt of Northeastern University called it significant and suggested it could influence future robotics, though he declined to predict a timeline, noting rapid changes in AI adoption recently.
The development also raises philosophical and ethical questions. If robots can self-correct and teach others, are they self-aware? Susan Schneider, who studies AI, warns that learning and self-correction don’t equal consciousness — the felt, inner quality of experience humans have. Still, she says the capabilities raise alarm bells for AI safety researchers: later versions might be misused or weaponized.
Researchers built safety protocols into the current systems, but they acknowledge the need for guardrails as the technology advances. Gupta and others suggest regulatory frameworks for who can operate such robots and how they’re used. As robots gain more adaptive, transferable skills, society faces choices about deployment, oversight and potential risks. It’s an exciting moment, but one with uncertainties about where these advances will lead.