On robot grasp learning using equivariant models
On robot grasp learning using equivariant models
复制标题
DOI：
10.1007/s10514-023-10112-w
复制
发表时间：
2023-06
影响因子：
3.5
通讯作者：
Xu Zhu;Dian Wang;Guanang Su;Ondrej Biza;R. Walters;Robert W. Platt
中科院分区：
文献类型：
--
作者：
Xu Zhu;Dian Wang;Guanang Su;Ondrej Biza;R. Walters;Robert W. Platt
由于抓取动力学的随机性和硬件中的噪声，现实世界的抓取检测具有挑战性。理想情况下，该系统将通过直接在物理系统上进行训练来适应真实的世界。然而，这通常是困难的，因为大多数抓取学习模型需要大量的训练数据。在本文中，我们注意到，平面把握功能是等变的，并证明了这种结构可以用来约束学习过程中使用的神经网络。这就产生了一种归纳偏差，可以显著提高抓取学习的样本效率，并在物理机器人上从头开始进行端到端的训练，只需600次抓取尝试。我们将这种方法称为对称抓取学习（SymGrasp），并表明它可以在不到1.5小时的物理机器人时间内“从头开始”学习抓取。本文是一个扩大和修订版的会议文件朱等。.
Real-world grasp detection is challenging due to the stochasticity in grasp dynamics and the noise in hardware. Ideally, the system would adapt to the real world by training directly on physical systems. However, this is generally difficult due to the large amount of training data required by most grasp learning models. In this paper, we note that the planar grasp function is-equivariant and demonstrate that this structure can be used to constrain the neural network used during learning. This creates an inductive bias that can significantly improve the sample efficiency of grasp learning and enable end-to-end training from scratch on a physical robot with as few as 600 grasp attempts. We call this method Symmetric Grasp learning (SymGrasp) and show that it can learn to grasp “from scratch” in less that 1.5 h of physical robot time. This paper represents an expanded and revised version of the conference paper Zhu et al. .