Robots Learn by Watching Humans

A team of scientists at École Polytechnique Fédérale de Lausanne in Switzerland has developed a method for robots to learn complex tasks by observing human demonstrations.

The approach uses machine learning and the robot's built-in awareness of its own body movement (kinematic intelligence) to copy and adapt human actions.

In demonstrations, robots watched a human tossing a ball into a container and then performed the same task while adjusting for their own position and design. The robots were able to transfer learned skills to other robots. The research was published in the journal Science Robotics.

A New Path for Robot Learning

Traditionally, robots have struggled to perform tasks beyond specific programming, especially in changing conditions. This new approach aims to enable robots to adapt to variations, similar to humans.

"I hope for a robot that could make me a coffee with specific customization."
— Sthithpragya Gupta, robotics scientist involved in the research

Robert Platt, a robotics researcher at Northeastern University not involved in the study, described the work as a potential "turning point" and a "breakthrough," but noted the field is not in agreement on the path forward.

The Limits of Machine Intelligence

Susan Schneider, an AI researcher at Florida Atlantic University, differentiated the robot's capabilities from consciousness, stating that robots lack the "felt quality of experience" that biological beings have.

Schneider raised concerns about potential future weaponization and safety risks, echoing the need for regulatory frameworks. Gupta acknowledged that future development will require guardrails and regulatory frameworks.

Safety First

Researchers have implemented safety protocols to prevent robots from harming people, though they recognize the need for ongoing oversight.