Projected Task-Specific Layers for Multi-Task Reinforcement Learning

Josselin Somerville Roberts, Julia Di
9/15/2023
cs.LGcs.AIcs.RO

Official code: JosselinSomervilleRoberts/PTSL (5 stars, MIT, last commit 2023-09-15).

Abstract

Multi-task reinforcement learning could enable robots to scale across a wide variety of manipulation tasks in homes and workplaces. However, generalizing from one task to another and mitigating negative task interference still remains a challenge. Addressing this challenge by successfully sharing information across tasks will depend on how well the structure underlying the tasks is captured. In this work, we introduce our new architecture, Projected Task-Specific Layers (PTSL), that leverages a common policy with dense task-specific corrections through task-specific layers to better express shared and variable task information. We then show that our model outperforms the state of the art on the MT10 and MT50 benchmarks of Meta-World consisting of 10 and 50 goal-conditioned tasks for a Sawyer arm.

View on arXivView PDF

Code Implementations(1)

JosselinSomervilleRoberts/PTSLOfficial100%
MIT

Repository of the paper: Projected Task-Specific Layers for Multi-Task Reinforcement Learning

50TeX, ShellMay 26, 20233 years agoMIT
machine-learningmetaworldptslreinforcement-learning-algorithmsrobotics

Similar Papers

Efficient Multi-Task Reinforcement Learning via Task-Specific Action Correction

86% match2024

T3S: Improving Multi-Task Reinforcement Learning with Task-Specific Feature Selector and Scheduler

86% match1 repo32026

Efficient Multi-task Reinforcement Learning with Cross-Task Policy Guidance

83% match1 repo92025

Rep-MTL: Unleashing the Power of Representation-level Task Saliency for Multi-Task Learning

82% match2025

Toward Task Generalization via Memory Augmentation in Meta-Reinforcement Learning

82% match2025

CompoSuite: A Compositional Reinforcement Learning Benchmark

82% match1 repo512022

Cite this paper

@article{roberts2023projected,
  title  = {Projected Task-Specific Layers for Multi-Task Reinforcement Learning},
  author = {Josselin Somerville Roberts and Julia Di},
  year   = {2023},
  eprint = {2309.08776},
  archivePrefix = {arXiv},
  url    = {http://arxiv.org/abs/2309.08776v2}
}

Add a badge to your README

Shows whether official code is linked for this paper.

Code of Paper badge
[![Code of Paper](https://codeofpaper.com/badge/2309.08776.svg)](https://codeofpaper.com/paper/2309.08776)

Discussion