When people work in pairs or teams, they can often solve a wider range of problems, completing some tasks faster and more efficiently than they would alone. To assist users similarly to how other humans would, robots should be able to rapidly interpret human behaviors and commands, using their predictions to plan and precisely execute helpful actions.

Researchers at GenerativeBionics, Italian Institute of Technology (IIT) and University of Manchester recently introduced a framework that could guide the design of robots that interact with humans more safely, efficiently and adaptively. Their design strategy, introduced in a paper published in Nature Machine Intelligence, was already used to create a new humanoid robot called ergoCub.

“Our work grew out of many years of experience developing different generations of humanoid robots,” Carlotta Sartore and Daniele Pucci, the paper’s first and senior authors, respectively, told Tech Xplore. “During this time, we repeatedly observed that robot hardware was often treated as fixed, while the control system was expected to compensate for its physical limitations.

Humanoids and humans are increasingly sharing factory floors, but one is made of metal and the other of flesh and bones, a mismatch that rarely ends well for us in a collision. Italian startup Generative Bionics thinks it has a fix: a humanoid robot covered in sensing skin that feels people approaching and adjusts its movements before any contact occurs.

The robot, called Gene.01, is wrapped in a network of sensors running from its torso to its limbs. This smart skin tracks touch, temperature, proximity, and force simultaneously, letting the robot anticipate a person’s presence and react before and during contact. It’s a bit like pulling your hand away from a hot stove before you actually touch it. The heat you feel from a distance is enough to make you stop.

The same sensors let humans physically teach the robot new tasks. Rather than only showing Gene.01 a movement on video, a person can guide its arm directly, helping it learn exactly how much force to apply. Generative Bionics says this solves one of humanoid robotics’ persistent headaches: teaching a machine to grip an object firmly enough that it doesn’t drop it but gently enough that it doesn’t crush it – a skill that’s hard to learn from video alone.