Addressing Exploration Challenges in Sparse Reward Reinforcement Learning Environments via Intrinsic Curiosity Modules and Reward Shaping

Fuente: Zenodo
Guardado en:
Detalles Bibliográficos
Autor principal: Elena Rossi
Formato: Recurso digital
Publicado: Zenodo 2026
Materias:
Acceso en línea:
Etiquetas: Agregar Etiqueta
Sin Etiquetas, Sea el primero en etiquetar este registro!
_version_ 1866901259967004672
author Elena Rossi
author_facet Elena Rossi
contents Reinforcement learning (RL) algorithms often struggle in environments with sparse rewards, leading to inefficient exploration and prolonged training times. This paper investigates the integration of intrinsic curiosity modules (ICM) with reward shaping techniques to mitigate these challenges. We propose a novel approach that combines ICM-generated intrinsic rewards with carefully designed extrinsic reward shaping functions to guide the agent's exploration and accelerate learning. Our experiments demonstrate the effectiveness of this combined approach in benchmark sparse reward environments, achieving significant improvements in sample efficiency and overall performance compared to traditional RL methods and ICM-only implementations.
format Recurso digital
id zenodo_https___doi_org_10_5281_zenodo_18919735
institution Zenodo
language
publishDate 2026
publisher Zenodo
record_format zenodo
spellingShingle Addressing Exploration Challenges in Sparse Reward Reinforcement Learning Environments via Intrinsic Curiosity Modules and Reward Shaping
Elena Rossi
machine learning
deep learning
artificial intelligence
Reinforcement learning (RL) algorithms often struggle in environments with sparse rewards, leading to inefficient exploration and prolonged training times. This paper investigates the integration of intrinsic curiosity modules (ICM) with reward shaping techniques to mitigate these challenges. We propose a novel approach that combines ICM-generated intrinsic rewards with carefully designed extrinsic reward shaping functions to guide the agent's exploration and accelerate learning. Our experiments demonstrate the effectiveness of this combined approach in benchmark sparse reward environments, achieving significant improvements in sample efficiency and overall performance compared to traditional RL methods and ICM-only implementations.
title Addressing Exploration Challenges in Sparse Reward Reinforcement Learning Environments via Intrinsic Curiosity Modules and Reward Shaping
topic machine learning
deep learning
artificial intelligence
url https://doi.org/10.5281/zenodo.18919735