Event-driven e-skin eases robots’ compute requirements

October 16, 2019 //By Julien Happich
e-skin
Aiming to give robots better proprioception and a better sense of their close surroundings, researchers at the Technical University of Munich (TUM) have developed a large-scale synthetic skin built up of 1260 tiled hexagonal cells, each containing a microprocessor and capable of sensing contact, acceleration, proximity and temperature.

With this large scale skin, they proceeded to almost entirely cover the body of an H-1 humanoid robot, with skin cells spreading on its upper body, arms, legs and even the soles of its feet. In effect, the sensitive feet also enables the robot to respond to uneven floor surfaces and even balance on one leg.


Prof. Gordon Cheng and his team developed a system
combining artificial skin with new control algorithms.
Image: Astrid Eckert / TUM

In order to reduce the processing power requirements of such a large number of sensors operating simultaneously, the researchers developed new event-based control algorithms directly inspired from neurosciences. The event-driven approach avoids data overload, as all the sensors no longer need to be interrogated simultaneously at precise clock-cycles but will only send out their data individually upon value changes.

The researchers reported their results in a paper titled “Evaluation of a Large Scale Event Driven Robot Skin” published in the IEEE Robotics and Automation Letters. The 1260 skin cells were distributed in 47 skin patches, combining in total 7560 multi-modal tactile sensors. The authors evaluated the efficiency of the robot skin in both a clock-driven mode and an event-based mode, demonstrating the superior performance of the event-driven system.


Vous êtes certain ?

Si vous désactivez les cookies, vous ne pouvez plus naviguer sur le site.

Vous allez être rediriger vers Google.