B站链接：https://www.bilibili.com/video/BV1n3b6zuExE/

开源代码：https://github.com/lalayh/trg

尽管深度学习模型在某些情况下可能实现成功的抓取，但它们往往难以准确反映特定抓取动作实际成功的可能性。为此本文在机器人抓取中引入了一个新的可信度问题，目的是弥合模型预测概率和实际抓取成功率的差距。本文提出了一种新的通过双分支网络组成的可信度对齐框架，有效提高了模型的可信度对齐性能同时也提高了抓取成功率和清除率。相比于可信度对齐之前，所提方法降低了期望抓取误差、最大抓取误差和过半抓取误差超过50%。我们相信这项工作能够为基于深度学习的现实世界机器人抓取奠定基础。

H. Yu, X. Zhang*, Z. Zhao, C. He, "Trustworthy Robotic Grasping: A Credibility Alignment Framework via Self-regulation Encoding", in IEEE/ASME Transactions on Mechatronics, doi: 10.1109/TMECH.2025.3598989

Abstract

Although deep learning models may achieve successful grasps in some instances, they often struggle to accurately reflect the true likelihood of success for a given grasp. In this paper, we introduce the trustworthy robotic grasping problem, aiming to bridge the gap between predicted grasp probabilities and actual grasp success rates. We propose a novel credibility alignment framework through a two-branch network architecture. This architecture generates an adjusting tensor for non-probabilistic outputs prior to the activation function of the backbone model, which is able to scale the output proportionally to improve the reliability of the predicted probability. To learn the adjusting tensor, a novel self-regulation encoder has been designed, which can extract 3D local features of the scene for the local associative regulation of non-probabilistic outputs. To facilitate research in this area, a new Trustworthy Robotic Grasping dataset has been created. Experimental results reveal that our method not only significantly reduces the expected grasp error, maximum grasp error, and latter half expected grasp error by up to 50% compared to the pre-credibility alignment state, but also enhances the grasp success and declutter rates. Real-world experiments further validate the efficacy of our method. Our code and dataset are available on https://github.com/lalayh/trg.