---
_id: '14220'
abstract:
- lang: eng
  text: Although reinforcement learning has seen remarkable progress over the last
    years, solving robust dexterous object-manipulation tasks in multi-object settings
    remains a challenge. In this paper, we focus on models that can learn manipulation
    tasks in fixed multi-object settings and extrapolate this skill zero-shot without
    any drop in performance when the number of objects changes. We consider the generic
    task of bringing a specific cube out of a set to a goal position. We find that
    previous approaches, which primarily leverage attention and graph neural network-based
    architectures, do not generalize their skills when the number of input objects
    changes while scaling as K2. We propose an alternative plug-and-play module based
    on relational inductive biases to overcome these limitations. Besides exceeding
    performances in their training environment, we show that our approach, which scales
    linearly in K, allows agents to extrapolate and generalize zero-shot to any new
    object number.
article_number: '2201.13388'
article_processing_charge: No
arxiv: 1
author:
- first_name: Davide
  full_name: Mambelli, Davide
  last_name: Mambelli
- first_name: Frederik
  full_name: Träuble, Frederik
  last_name: Träuble
- first_name: Stefan
  full_name: Bauer, Stefan
  last_name: Bauer
- first_name: Bernhard
  full_name: Schölkopf, Bernhard
  last_name: Schölkopf
- first_name: Francesco
  full_name: Locatello, Francesco
  id: 26cfd52f-2483-11ee-8040-88983bcc06d4
  last_name: Locatello
  orcid: 0000-0002-4850-0683
citation:
  ama: Mambelli D, Träuble F, Bauer S, Schölkopf B, Locatello F. Compositional multi-object
    reinforcement learning with linear relation networks. <i>arXiv</i>. doi:<a href="https://doi.org/10.48550/arXiv.2201.13388">10.48550/arXiv.2201.13388</a>
  apa: Mambelli, D., Träuble, F., Bauer, S., Schölkopf, B., &#38; Locatello, F. (n.d.).
    Compositional multi-object reinforcement learning with linear relation networks.
    <i>arXiv</i>. <a href="https://doi.org/10.48550/arXiv.2201.13388">https://doi.org/10.48550/arXiv.2201.13388</a>
  chicago: Mambelli, Davide, Frederik Träuble, Stefan Bauer, Bernhard Schölkopf, and
    Francesco Locatello. “Compositional Multi-Object Reinforcement Learning with Linear
    Relation Networks.” <i>ArXiv</i>, n.d. <a href="https://doi.org/10.48550/arXiv.2201.13388">https://doi.org/10.48550/arXiv.2201.13388</a>.
  ieee: D. Mambelli, F. Träuble, S. Bauer, B. Schölkopf, and F. Locatello, “Compositional
    multi-object reinforcement learning with linear relation networks,” <i>arXiv</i>.
    .
  ista: Mambelli D, Träuble F, Bauer S, Schölkopf B, Locatello F. Compositional multi-object
    reinforcement learning with linear relation networks. arXiv, 2201.13388.
  mla: Mambelli, Davide, et al. “Compositional Multi-Object Reinforcement Learning
    with Linear Relation Networks.” <i>ArXiv</i>, 2201.13388, doi:<a href="https://doi.org/10.48550/arXiv.2201.13388">10.48550/arXiv.2201.13388</a>.
  short: D. Mambelli, F. Träuble, S. Bauer, B. Schölkopf, F. Locatello, ArXiv (n.d.).
date_created: 2023-08-22T14:23:16Z
date_published: 2022-01-31T00:00:00Z
date_updated: 2024-10-14T12:27:39Z
day: '31'
department:
- _id: FrLo
doi: 10.48550/arXiv.2201.13388
extern: '1'
external_id:
  arxiv:
  - '2201.13388'
language:
- iso: eng
main_file_link:
- open_access: '1'
  url: https://doi.org/10.48550/arXiv.2201.13388
month: '01'
oa: 1
oa_version: Preprint
publication: arXiv
publication_status: submitted
status: public
title: Compositional multi-object reinforcement learning with linear relation networks
type: preprint
user_id: 2DF688A6-F248-11E8-B48F-1D18A9856A87
year: '2022'
...