Implicit Kinematic Policies: Unifying Joint and Cartesian Action Spaces in End-to-End Robot Learning

被引:4
|
作者
Ganapathi, Aditya [1 ,2 ]
Florence, Pete [1 ]
Varley, Jake [1 ]
Burns, Kaylee [1 ,3 ]
Goldberg, Ken [2 ]
Zeng, Andy [1 ]
机构
[1] Google, Robot, Mountain View, CA 94043 USA
[2] Univ Calif Berkeley, Berkeley, CA 90095 USA
[3] Stanford Univ, Stanford, CA USA
来源
2022 IEEE INTERNATIONAL CONFERENCE ON ROBOTICS AND AUTOMATION (ICRA 2022) | 2022年
关键词
D O I
10.1109/ICRA46639.2022.9812165
中图分类号
TP [自动化技术、计算机技术];
学科分类号
0812 ;
摘要
Action representation is an important yet often overlooked aspect in end-to-end robot learning with deep networks. Choosing one action space over another (e.g. target joint positions, or Cartesian end-effector poses) can result in surprisingly stark performance differences between various downstream tasks - and as a result, considerable research has been devoted to finding the right action space for a given application. However, in this work, we instead investigate how our models can discover and learn for themselves which action space to use. Leveraging recent work on implicit behavioral cloning, which takes both observations and actions as input, we demonstrate that it is possible to present the same action in multiple different spaces to the same policy - allowing it to learn inductive patterns from each space. Specifically, we study the benefits of combining Cartesian and joint action spaces in the context of learning manipulation skills. To this end, we present Implicit Kinematic Policies (IKP), which incorporates the kinematic chain as a differentiable module within the deep network. Quantitative experiments across several simulated continuous control tasks-from scooping piles of small objects, to lifting boxes with elbows, to precise block insertion with miscalibrated robots-suggest IKP not only learns complex prehensile and non-prehensile manipulation from pixels better than baseline alternatives, but also can learn to compensate for small joint encoder offset errors. Finally, we also run qualitative experiments on a real UR5e to demonstrate the feasibility of our algorithm on a physical robotic system with real data. See https://tinyurl.com/4wz3nf86 for code and supplementary material.
引用
收藏
页码:2656 / 2662
页数:7
相关论文
共 50 条
  • [41] Rank pooling dynamic network: Learning end-to-end dynamic characteristic for action recognition
    Zhu, Zhigang
    Ji, Hongbing
    Zhang, Wenbo
    Xu, Yiping
    NEUROCOMPUTING, 2018, 317 : 101 - 109
  • [42] An End-to-End Calibration Method for Welding Robot Laser Vision Systems With Deep Reinforcement Learning
    Zou, Yanbiao
    Lan, Rui
    IEEE TRANSACTIONS ON INSTRUMENTATION AND MEASUREMENT, 2020, 69 (07) : 4270 - 4280
  • [43] XBG: End-to-End Imitation Learning for Autonomous Behaviour in Human-Robot Interaction and Collaboration
    Cardenas-Perez, Carlos
    Romualdi, Giulio
    Elobaid, Mohamed
    Dafarra, Stefano
    L'Erario, Giuseppe
    Traversaro, Silvio
    Morerio, Pietro
    Del Bue, Alessio
    Pucci, Daniele
    IEEE ROBOTICS AND AUTOMATION LETTERS, 2024, 9 (12): : 11617 - 11624
  • [44] Gershgorin Loss Stabilizes the Recurrent Neural Network Compartment of an End-to-end Robot Learning Scheme
    Lechner, Mathias
    Hasani, Ramin
    Rus, Daniela
    Grosu, Radu
    2020 IEEE INTERNATIONAL CONFERENCE ON ROBOTICS AND AUTOMATION (ICRA), 2020, : 5446 - 5452
  • [45] An End-To-End Earthquake Detection Method for Joint Phase Picking and Association Using Deep Learning
    Zhu, Weiqiang
    Tai, Kai Sheng
    Mousavi, S. Mostafa
    Bailis, Peter
    Beroza, Gregory C.
    JOURNAL OF GEOPHYSICAL RESEARCH-SOLID EARTH, 2022, 127 (03)
  • [46] FCSR-GAN: End-to-end Learning for Joint Face Completion and Super-resolution
    Cai, Jiancheng
    Hu, Han
    Shan, Shiguang
    Chen, Xilin
    2019 14TH IEEE INTERNATIONAL CONFERENCE ON AUTOMATIC FACE AND GESTURE RECOGNITION (FG 2019), 2019, : 16 - 23
  • [47] End-to-end deep learning for joint geometric-probabilistic constellation shaping in FMF system
    Amirabadi, Mohammad Ali
    Kahaei, Mohammad Hossein
    Nezamalhosseini, S. Alireza
    PHYSICAL COMMUNICATION, 2022, 55
  • [48] An end-to-end joint learning scheme of image compression and quality enhancement with improved entropy minimization
    Lee, Jooyoung
    Cho, Seunghyun
    Kim, Munchurl
    ETRI JOURNAL, 2024, 46 (06) : 935 - 949
  • [49] Signal Enhancement Aided End-to-End Deep Learning Approach for Joint Denoising and Spectrum Sensing
    Su, Zhengyang
    Teh, Kah Chan
    Xie, Yihang
    Razul, Sirajudeen Gulam
    Kot, Alex C.
    IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, 2024, 73 (03) : 4424 - 4428
  • [50] LkeRec: Toward Lightweight End-to-End Joint Representation Learning for Building Accurate and Effective Recommendation
    Yan, Surong
    Lin, Kwei-Jay
    Zheng, Xiaolin
    Wang, Haosen
    ACM TRANSACTIONS ON INFORMATION SYSTEMS, 2022, 40 (03)