Robots could negotiate awkward terrain surefootedly thanks to squidgy paws containing cameras.
at the Indian Institute of Technology Bombay, at the Norwegian University of Science and Technology and their colleagues have developed what they call a Terrain Recognition and Contact Force Estimation Paw ().
The bottom part of the foot is half a silicone ball, which deforms as the robot walks. Inside the ball, a camera takes images of a grid-like pattern of dots on the interior of the silicone, detecting how it changes in response to each step. The images are sent to an internal AI model, which works out what force is being applied and from what direction.
Advertisement
A microphone also captures the sound made by each step and another AI model uses this to differentiate between surfaces like gravel, snow, grass and sand to give the robot a greater understanding of its location and potentially more stable locomotion.
鈥淚f you can detect the soil type, and then give this information to the robot, it can adjust the way it walks; just as you would walk differently if you walked on concrete, gravel, or sand, or ice,鈥� says Olsen.
A small computer chip inside the foot can run all the AI algorithms, so extensive wiring isn鈥檛 needed throughout the robot. The 2-megapixel camera and microphone are off-the-shelf components available cheaply 鈥� the entire electronics package costs around 拢100 and is about the size of a 拢1 coin but just 1 millimetre thick.
To do the same thing, a robot would conventionally use a range of force and torque sensors. In tests, the researchers found that TRACEPaw was almost as accurate as force and torque sensors, but crucially cheaper and less bulky.
The researchers tested TRACEPaw as an individual leg but Olsen says a robot could have a smart foot on each leg, sending information back to a central computer, or just on one leg, which is used as a sort of probe.
Barnwal says TRACEPaw essentially gives robots more senses. 鈥淎s humans, we have the sense of touch, a sense of sight and a sense of sound, and using all these three senses we鈥檙e actually able to walk in any kind of adversity or difficulty. So empowering a legged robot with these kinds of sensing would also help in a very similar manner,鈥� she says.
arXiv
