maze-2d

Maze2D is a benchmark dataset used to evaluate exploration strategies in visual reinforcement learning, particularly in environments characterized by sparse rewards and complex task-relevant variations.

Papers using maze-2d (4)

Task-Aware Exploration via a Predictive Bisimulation Metric2026Generative Actor Critic2025DIAR: Diffusion-model-guided Implicit Q-learning with Adaptive Revaluation2024Active Reinforcement Learning Strategies for Offline Policy Improvement2024