BriefGPT.xyz
Feb, 2023
通过修剪原型目标扩展目标导向探索的规模
Scaling Goal-based Exploration via Pruning Proto-goals
HTML
PDF
Akhil Bagaria, Ray Jiang, Ramana Kumar, Tom Schaul
TL;DR
本研究基于强化学习，通过在人工设计产生的广泛目标空间中寻找可控、可达、新颖和相关目标的自主发现过程，弥补探索广域领域中新奇性和涵盖性行为不足的问题，并在三种具有挑战性的环境中证明了目标导向的探索的有效性。
Abstract
One of the gnarliest challenges in
reinforcement learning
(RL) is
exploration
that scales to vast domains, where novelty-, or coverage-seeking behaviour falls short. Goal-directed, purposeful behaviours are able
→