Vision-based tactile sensing with DIGIT

A few pictures and schemes of the lab work of Ronan Lawlor during his MEng Project.

How can optical-based tactile sensing be implemented in a computationally efficient manner, extracting only the necessary information? The proposed solution to this these problems is to replace the RGB cameras typically used in optical tactile sensing with event cameras. These take a bio-inspired approach to image capture, only sampling pixels which detect movement more than a set threshold and so mitigating the high data rate present even in the absence of contact. In order to take full advantage of the camera’s improved properties, new algorithms must be developed which are both fast and computationally inexpensive. The algorithms developed in this project are not limited to event cameras and can instead be applied to any suitable tactile sensor. 

The DIGIT sensor

(Emulated) event-based readout

Tracking

Texture classification