Uncertainty-Aware Knowledge Distillation for Collision Identification of Collaborative Robots

Abstract

Human-robot interaction has received a lot of attention as collaborative robots became widely utilized in many industrial fields. Among techniques for human-robot interaction, collision identification is an indispensable element in collaborative robots to prevent fatal accidents. This paper proposes a deep learning method for identifying external collisions in 6-DoF articulated robots. The proposed method expands the idea of CollisionNet, which was previously proposed for collision detection, to identify the locations of external forces. The key contribution of this paper is uncertainty-aware knowledge distillation for improving the accuracy of a deep neural network. Sample-level uncertainties are estimated from a teacher network, and larger penalties are imposed for uncertain samples during the training of a student network. Experiments demonstrate that the proposed method is effective for improving the performance of collision identification.

Automating mosquito taxonomy by compressing and enhancing a feature fused EfficientNet with knowledge distillation and a novel residual skip block

“…To produce the Hard Loss in (12), the student model takes the predictions P and maps it with ground truth labels G produced from the modified softmax τ>1. The Hard Loss uses a standard categorical-cross entropy loss CCE Loss [34] with labels set to >2, as the proposed method has six classes. …”

Human–Robot Collaboration in Industrial Automation: Sensors and Algorithms

“…These engineers will also need to design connected systems that can efficiently and safely interact with humans during the manufacturing process, e.g., a car assembly line [ 3 ]. The focal points of this Special Issue are the smart sensors that enable robots and humans to “see” each other [ 4 , 5 , 6 , 7 , 8 , 9 ] and the machine learning algorithms that process these complex data so the robot can make decisions [ 10 , 11 , 12 , 13 ].…”

“…A neural network model has been previously developed to determine when a collision has occurred [ 17 ] so the robot can adjust its force and avoid an accident. Kwon expanded this neural network to include where on the robot the collision occurred [ 11 ]. This work is important for safety, especially as robots become more complicated with more articulations.…”