arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2103.14633cs.ROcs.CVcs.LGcs.NE
收藏

Visionary:面向机器人学习的视觉架构发现

Visionary: Vision architecture discovery for robot learning

Robotics at Google(谷歌机器人部门);Stony Brook University(石溪大学);Columbia University(哥伦比亚大学)

Iretiayo Akinola, Anelia Angelova, Yao Lu, Yevgen Chebotar, Dmitry Kalashnikov, Jacob Varley, Julian Ibarz, Michael S. Ryoo

摘要:

提出了一种基于视觉的机器人操作学习体系结构搜索算法,该算法能够发现低维动作输入和高维视觉输入之间的相互作用。我们的方法在训练任务的同时自动设计架构-发现将图像特征表示与动作以及先前层的特征相结合和参与的新方法。所获得的新架构表现出更好的任务成功率,在某些情况下,与最近的高性能基线相比,具有很大的利润。我们的真实机器人实验也证实,它提高了6%的抓取性能。这是第一种方法来证明一个成功的神经架构搜索和注意力连接搜索一个真正的机器人任务。

英文摘要:

We propose a vision-based architecture search algorithm for robot manipulation learning, which discovers interactions between low dimension action inputs and high dimensional visual inputs. Our approach automatically designs architectures while training on the task - discovering novel ways of combining and attending image feature representations with actions as well as features from previous layers. The obtained new architectures demonstrate better task success rates, in some cases with a large margin, compared to a recent high performing baseline. Our real robot experiments also confirm that it improves grasping performance by 6%. This is the first approach to demonstrate a successful neural architecture search and attention connectivity search for a real-robot task.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑