Imagine a robot that folds your laundry, makes your bed, cooks dinner or restocks milk at the store. For decades engineers have been able to teach robots single, fixed tasks, but getting machines to handle complex, changing jobs has proven difficult despite large investments in robotics.
A research team in Switzerland reports progress toward robots that can follow richer human instructions. The advance expands possibilities for useful helpers, and it also raises questions about whether such machines could be misused.
Sthithpragya Gupta, a robotics researcher at École Polytechnique Fédérale de Lausanne, says he dreams of a robot that can make his coffee — one he could tell ‘a little bit of sugar, a bit more creamer.’ The core technical obstacle is adaptability. Robots can be trained to repeat a precise motion reliably, but small variations in the environment or task often break those behaviors. Humans adjust on the fly; giving robots that flexibility has been a major challenge.
Gupta and colleagues published a paper in Science Robotics describing a machine-learning approach that relies on what they call kinematic intelligence — an internal model of how a robot’s own body can move safely through space. In videos demonstrating the method, single-arm robots watch a person toss a ball into a small container. The robots then pick up a ball and imitate the motion while compensating for different starting positions and nonhuman limbs. The resulting skill can be transferred to other robots.
Robert Platt, a robotics researcher at Northeastern University, called the work a breakthrough that tackles a critical problem in robot learning. He also warned that predicting when these capabilities will appear in everyday settings is difficult; technological shifts can accelerate rapidly, as recent progress in large language models showed.
If robots can self-correct and teach each other, does that amount to self-awareness? Many scholars say no. Susan Schneider, who studies AI at Florida Atlantic University, emphasizes that such systems can learn impressively but do not have felt experience. She notes that consciousness involves the inner quality of experience — the sensation of drinking an espresso or seeing a sunset — which current machines lack.
Still, the research brings ethical and safety concerns. Schneider warns that future versions could be weaponized. The Swiss team has implemented safety protocols to reduce risk, but they acknowledge that broader guardrails will be needed. Gupta urges regulatory frameworks to define who can operate robots and how they may be used.
Humans stand at an inflection point with robotics. The work is promising and could enable helpful assistants, but its ultimate direction — beneficial or dangerous — remains uncertain and will depend on technical safeguards, regulation, and how society chooses to deploy these systems.