The "Terrain Recognition And Contact force Estimation Paw" (TRACE Paw) is a compact and sensorized "smart" paw for robot dogs and other legged robots enabling a) vision-based force estimation and b) audio-based terrain classification on the edge.

Through silicon deformation on the end effector combined with an embedded micro camera, and a microphone, alongside the necessary compute fully integrated in the paw, the design exploits advancements in neural networks and real-time inference to a) map silicon deformation observed inside the paw to a force estimate and b) the distinct sound features produced from the interaction with the terrain to infer if the robot is on snow, grass, concrete and other types of surfaces.

We release the complete mechatronic design, datasets, ML models and full software for the system.

Repository: https://github.com/ntnu-arl/trace_paw

The Autonomous Robots Lab
https://www.autonomousrobom.com