Exploration-Intensive Distractors: Two Environment Proposals and a Benchmarking
Abstract
Sparse-reward environments are famously challenging for deep reinforcement learning (DRL) algorithms. Yet, the prospect of solving intrinsically sparse tasks in an end-to-end fashion without any extra reward engineering is highly appealing. Such aspiration has recently led to the development of numerous DRL algorithms able to handle sparse-reward environments to some extent. Some methods have even gone one step further and have tackled sparse tasks involving different kinds of distractors (e.g. broken TV, self-moving phantom objects and many more). In this work, we put forward two motivating new sparse-reward environments containing the so-far largely overlooked class of exploration-intensive distractors. Furthermore, we conduct a benchmarking which reveals that state-of-the-art algorithms are not yet all-around suitable for solving the proposed environments.
Venue
AIxIA 2021
Date
2021