DreamSmooth: Improving Model-based Reinforcement Learning via Reward Smoothing

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Lee, Vint, Abbeel, Pieter, Lee, Youngwoon
Format: Preprint
Published: 2023
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866910334444371968
author Lee, Vint
Abbeel, Pieter
Lee, Youngwoon
author_facet Lee, Vint
Abbeel, Pieter
Lee, Youngwoon
contents Model-based reinforcement learning (MBRL) has gained much attention for its ability to learn complex behaviors in a sample-efficient way: planning actions by generating imaginary trajectories with predicted rewards. Despite its success, we found that surprisingly, reward prediction is often a bottleneck of MBRL, especially for sparse rewards that are challenging (or even ambiguous) to predict. Motivated by the intuition that humans can learn from rough reward estimates, we propose a simple yet effective reward smoothing approach, DreamSmooth, which learns to predict a temporally-smoothed reward, instead of the exact reward at the given timestep. We empirically show that DreamSmooth achieves state-of-the-art performance on long-horizon sparse-reward tasks both in sample efficiency and final performance without losing performance on common benchmarks, such as Deepmind Control Suite and Atari benchmarks.
format Preprint
id arxiv_https___arxiv_org_abs_2311_01450
institution arXiv
publishDate 2023
record_format arxiv
spellingShingle DreamSmooth: Improving Model-based Reinforcement Learning via Reward Smoothing
Lee, Vint
Abbeel, Pieter
Lee, Youngwoon
Machine Learning
Artificial Intelligence
Robotics
Model-based reinforcement learning (MBRL) has gained much attention for its ability to learn complex behaviors in a sample-efficient way: planning actions by generating imaginary trajectories with predicted rewards. Despite its success, we found that surprisingly, reward prediction is often a bottleneck of MBRL, especially for sparse rewards that are challenging (or even ambiguous) to predict. Motivated by the intuition that humans can learn from rough reward estimates, we propose a simple yet effective reward smoothing approach, DreamSmooth, which learns to predict a temporally-smoothed reward, instead of the exact reward at the given timestep. We empirically show that DreamSmooth achieves state-of-the-art performance on long-horizon sparse-reward tasks both in sample efficiency and final performance without losing performance on common benchmarks, such as Deepmind Control Suite and Atari benchmarks.
title DreamSmooth: Improving Model-based Reinforcement Learning via Reward Smoothing
topic Machine Learning
Artificial Intelligence
Robotics
url https://arxiv.org/abs/2311.01450