Researchers at the Massachusetts Institute of Technology (https://www.mit.edu/ ) (MIT) and the University of British Columbia (https://www.ubc.ca/) are enabling robots to identify objects based on their weight and mass distribution if there is insufficient light to recognise an object visually. The researchers are thinking, for example, of searching for people in disaster areas, where it is necessary to distinguish whether the object found is a stone or a human being – without shaking it. In this case, sensors are activated to check the surface texture and hardness of the object.

Stress on joints is relevant

The method developed by the international research team, which also includes developers from Amazon (https://www.amazon.com), uses proprioception. This refers to the ability of humans or robots to perceive their movement or position in space. For example, when a person lifts a dumbbell in the gym, they feel the weight in their wrist and biceps. In the same way, a robot can ‘feel’ the weight of an object by the load on the joints in its arm.

When the robot lifts an object, the system collects signals from the robot’s rotary encoders, i.e. sensors that detect the direction and speed of rotation of the joints. ‘Most robots have such sensors,’ says Chao Liu, a postdoctoral researcher in computer science and artificial intelligence at MIT. ’This makes our technology inexpensive because it does not require additional components such as tactile sensors or image tracking systems.’

‘Computer vision’ remains important

‘We don’t want to replace what is known as ‘computer vision’,’ emphasises Peter Yichen Chen, a postdoctoral researcher at MIT who earned his PhD in computer science at the University of British Columbia. “Both methods have their advantages and disadvantages. But we have shown that we can already determine some properties of objects without a camera,” the scientist concludes.