Evaluating Long-Term Memory in 3D Mazes

TL;DR

This paper introduces the Memory Maze, a 3D environment designed to evaluate long-term memory in reinforcement learning agents, highlighting current algorithms' limitations and providing benchmarks for future improvements.

Contribution

The paper presents the Memory Maze environment, a new benchmark for assessing long-term memory in agents, along with datasets, evaluation methods, and baseline comparisons.

Findings

Current algorithms perform well on small mazes but struggle with large mazes.

Training with truncated backpropagation through time improves performance.

Humans outperform algorithms on complex maze navigation tasks.

Abstract

Intelligent agents need to remember salient information to reason in partially-observed environments. For example, agents with a first-person view should remember the positions of relevant objects even if they go out of view. Similarly, to effectively navigate through rooms agents need to remember the floor plan of how rooms are connected. However, most benchmark tasks in reinforcement learning do not test long-term memory in agents, slowing down progress in this important research direction. In this paper, we introduce the Memory Maze, a 3D domain of randomized mazes specifically designed for evaluating long-term memory in agents. Unlike existing benchmarks, Memory Maze measures long-term memory separate from confounding agent abilities and requires the agent to localize itself by integrating information over time. With Memory Maze, we propose an online reinforcement learning benchmark, a diverse offline dataset, and an offline probing evaluation. Recording a human player establishes a strong baseline and verifies the need to build up and retain memories, which is reflected in their gradually increasing rewards within each episode. We find that current algorithms benefit from training with truncated backpropagation through time and succeed on small mazes, but fall short of human performance on the large mazes, leaving room for future algorithmic designs to be evaluated on the Memory Maze.