As previous representations for reinforcement learning cannot effectively incorporate a human-intuitive understanding of the 3D environment, they usually suffer from sub-optimal performances. In this paper, we present Semantic-aware Neural Radiance Fields for Reinforcement Learning (SNeRL), which jointly optimizes semantic-aware neural radiance fields (NeRF) with a convolutional encoder to learn 3D-aware neural implicit representation from multi-view images. We introduce 3D semantic and distilled feature fields in parallel to the RGB radiance fields in NeRF to learn semantic and object-centric representation for reinforcement learning. SNeRL outperforms not only previous pixel-based representations but also recent 3D-aware representations both in model-free and model-based reinforcement learning.
@article{arxiv.2301.11520,
title = {SNeRL: Semantic-aware Neural Radiance Fields for Reinforcement Learning},
author = {Dongseok Shim and Seungjae Lee and H. Jin Kim},
journal= {arXiv preprint arXiv:2301.11520},
year = {2023}
}
Comments
ICML 2023. First two authors contributed equally. Order was determined by coin flip