AnyMorph: Learning Transferable Polices By Inferring Agent Morphology

被引:0
|
作者
Trabucco, Brandon [1 ,2 ]
Phielipp, Mariano [2 ]
Berseth, Glen [3 ]
机构
[1] Carnegie Mellon Univ, Machine Learning Dept, Pittsburgh, PA 15213 USA
[2] Intel AI, San Diego, CA USA
[3] Mila, Montreal, PQ, Canada
关键词
D O I
暂无
中图分类号
TP18 [人工智能理论];
学科分类号
081104 ; 0812 ; 0835 ; 1405 ;
摘要
The prototypical approach to reinforcement learning involves training policies tailored to a particular agent from scratch for every new morphology. Recent work aims to eliminate the re-training of policies by investigating whether a morphology-agnostic policy, trained on a diverse set of agents with similar task objectives, can be transferred to new agents with unseen morphologies without re-training. This is a challenging problem that required previous approaches to use hand-designed descriptions of the new agent's morphology. Instead of hand-designing this description, we propose a data-driven method that learns a representation of morphology directly from the reinforcement learning objective. Ours is the first reinforcement learning algorithm that can train a policy to generalize to new agent morphologies without requiring a description of the agent's morphology in advance. We evaluate our approach on the standard benchmark for agent-agnostic control, and improve over the current state of the art in zero-shot generalization to new agents. Importantly, our method attains good performance without an explicit description of morphology.
引用
收藏
页数:15
相关论文
共 50 条
  • [1] Inferring transferable intermolecular potential models
    Ucyigitler, Sinan
    Camurdan, Mehmet C.
    Turkay, Metin
    Elliott, J. Richard
    MOLECULAR SIMULATION, 2008, 34 (02) : 147 - 154
  • [2] An Analysis of Polices in Terms of Lifelong Learning
    张守兵
    罗佳
    英语广场(学术研究), 2012, (05) : 81 - 82
  • [3] Efficient Exploration for Multi-Agent Reinforcement Learning via Transferable Successor Features
    Wenzhang Liu
    Lu Dong
    Dan Niu
    Changyin Sun
    IEEE/CAA Journal of Automatica Sinica, 2022, 9 (09) : 1673 - 1686
  • [4] Efficient Exploration for Multi-Agent Reinforcement Learning via Transferable Successor Features
    Liu, Wenzhang
    Dong, Lu
    Niu, Dan
    Sun, Changyin
    IEEE-CAA JOURNAL OF AUTOMATICA SINICA, 2022, 9 (09) : 1673 - 1686
  • [5] Inferring the goals of a nonhuman agent
    Johnson, SC
    Booth, A
    O'Hearn, K
    COGNITIVE DEVELOPMENT, 2001, 16 (01) : 637 - 656
  • [6] Learning a Transferable World Model by Reinforcement Agent in Deterministic Observable Grid-World Environments
    Kapociute-Dzikiene, Jurgita
    Raskinis, Gailius
    INFORMATION TECHNOLOGY AND CONTROL, 2012, 41 (04): : 318 - 327
  • [7] Learning theory for inferring interaction kernels in second-order interacting agent systems
    Miller, Jason
    Tang, Sui
    Zhong, Ming
    Maggioni, Mauro
    SAMPLING THEORY SIGNAL PROCESSING AND DATA ANALYSIS, 2023, 21 (01):
  • [8] Transferable learning on analog hardware
    Vadlamani, Sri Krishna
    Englund, Dirk
    Hamerly, Ryan
    SCIENCE ADVANCES, 2023, 9 (28)
  • [9] TRANSFERABLE SKILLS AND LIFELONG LEARNING
    Endrawes, Gihane
    4TH INTERNATIONAL TECHNOLOGY, EDUCATION AND DEVELOPMENT CONFERENCE (INTED 2010), 2010, : 5188 - 5194
  • [10] Learning to Learn Transferable Attack
    Fang, Shuman
    Li, Jie
    Lin, Xianming
    Ji, Rongrong
    THIRTY-SIXTH AAAI CONFERENCE ON ARTIFICIAL INTELLIGENCE / THIRTY-FOURTH CONFERENCE ON INNOVATIVE APPLICATIONS OF ARTIFICIAL INTELLIGENCE / THE TWELVETH SYMPOSIUM ON EDUCATIONAL ADVANCES IN ARTIFICIAL INTELLIGENCE, 2022, : 571 - 579