IROS 2016 "Best Paper on Cognitive Robotics Finalist” and “Best Student Paper Finalist”
In this video we show that material classification purely based on the spatio-temporal signal of a flexible tactile skin mounted on the finger tip of the advanced humanoid robot Agile Justin can be robustly performed in a real world setting. We develop a convolutional deep learning network architecture which is directly fed with the raw 24000 dimensional sensor signal of the tactile skin. The network with its 16 million weights is trained from only 540 samples and reaches a classification accuracy of up to 97.3%.
S. Baishya and B. Bäuml. Robust material classification with a tactile skin using deep learning. In Proc. IEEE International Conference on Intelligent Robots and Systems, 2016.
Agile Justin is DLR's advanced humanoid robot for research in autonomous learning (see • Agile Justin: An Upgra... for details).