This paper proposes a new network, the Modularized Attention-Dilated Convolutional Neural Network (MAD-CNN), specifically designed for collision detection in robots equipped with variable stiffness actuators. The authors aim to enhance safety in the field of collaborative robotics. MAD-CNN incorporates a dual inductive bias mechanism and an attention module to enhance data efficiency and improve robustness. Despite limited training data, MAD-CNN effectively detects all collisions with minimal delay across various stiffness conditions. It outperforms existing models in terms of collision sensitivity and robustness.
Publication date: 5 Oct 2023
Project Page: https://arxiv.org/abs/2310.02573v1
Paper: https://arxiv.org/pdf/2310.02573
