Minihack

MiniHack is a benchmark dataset that contains a collection of reinforcement learning environments designed to evaluate the performance of algorithms in scenarios with state-dependent action validity.

Papers using Minihack (7)

Overcoming Valid Action Suppression in Unmasked Policy Gradient Algorithms2026Robust Agents in Open-Ended Worlds2025Scalable Option Learning in High-Throughput Environments2025Growing with Experience: Growing Neural Networks in Deep Reinforcement Learning2025CAE: Repurposing the Critic as an Explorer in Deep Reinforcement Learning2025Efficient Exploration and Discriminative World Model Learning with an Object-Centric Abstraction2024Assessing the Zero-Shot Capabilities of LLMs for Action Evaluation in RL2024