DrS: Learning Reusable Dense Rewards for Multi-Stage Tasks

Tongzhou Mu, Minghua Liu, Hao Su
4/25/2024
cs.LGcs.AIcs.RO

No official code found. Best implementation: tongzhoumu/DrS (22 stars, MIT, last commit 2024-04-26).

Abstract

The success of many RL techniques heavily relies on human-engineered dense rewards, which typically demand substantial domain expertise and extensive trial and error. In our work, we propose DrS (Dense reward learning from Stages), a novel approach for learning reusable dense rewards for multi-stage tasks in a data-driven manner. By leveraging the stage structures of the task, DrS learns a high-quality dense reward from sparse rewards and demonstrations if given. The learned rewards can be \textit{reused} in unseen tasks, thus reducing the human effort for reward engineering. Extensive experiments on three physical robot manipulation task families with 1000+ task variants demonstrate that our learned rewards can be reused in unseen tasks, resulting in improved performance and sample efficiency of RL algorithms. The learned rewards even achieve comparable performance to human-engineered rewards on some tasks. See our project page (https://sites.google.com/view/iclr24drs) for more details.

View on arXivView PDF

Code Implementations(1)

tongzhoumu/DrSCommunity72%
MIT

Code for "DrS: Learning Reusable Dense Rewards for Multi-Stage Tasks"

222PythonApr 22, 20242 years agoMIT
reinforcement-learningreward-shapingrobotics

Similar Papers

Towards better dense rewards in Reinforcement Learning Applications

84% match2025

Multi-Stage Manipulation with Demonstration-Augmented Reward, Policy, and World Model Learning

83% match1 repo682025

Auto MC-Reward: Automated Dense Reward Design with Large Language Models for Minecraft

82% match2026

Image-Based Deep Reinforcement Learning with Intrinsically Motivated Stimuli: On the Execution of Complex Robotic Tasks

82% match1 repo92024

Open Problems and Modern Solutions for Deep Reinforcement Learning

82% match2023

On-Robot Reinforcement Learning with Goal-Contrastive Rewards

82% match2024

Cite this paper

@article{mu2024drs,
  title  = {DrS: Learning Reusable Dense Rewards for Multi-Stage Tasks},
  author = {Tongzhou Mu and Minghua Liu and Hao Su},
  year   = {2024},
  eprint = {2404.16779},
  archivePrefix = {arXiv},
  url    = {http://arxiv.org/abs/2404.16779v1},
  journal = {ICLR 2024 2024}
}

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/2404.16779.svg)](https://codeofpaper.com/paper/2404.16779)

Discussion