AntMaze

AntMaze is a benchmark within the D4RL continuous-control dataset that is used to evaluate offline reinforcement learning algorithms by providing a set of maze navigation tasks for agents.

Papers using AntMaze (17)

Counterfactual Transport Flows for Offline Conservative Trajectory Refinement2026When Policies Cannot Be Retrained: A Unified Closed-Form View of Post-Training Steering in Offline Reinforcement Learning2026Inference Time Policy Optimization for Offline RL with Differentiable World Models2026Offline-to-Online Reinforcement Learning with Classifier-Free Diffusion Generation2025Online Pre-Training for Offline-to-Online Reinforcement Learning2025Variational OOD State Correction for Offline Reinforcement Learning2025MRS: Multi-Resolution Skills for HRL Agents2025Behavior Preference Regression for Offline Reinforcement Learning2025Predictive Coding for Decision Transformer2024 · 3 citesDoubly Mild Generalization for Offline Reinforcement Learning2024 · 1 citesAlignIQL: Policy Alignment in Implicit Q-Learning through Constrained Optimization2024Preferred-Action-Optimized Diffusion Policies for Offline Reinforcement Learning2024PlanDQ: Hierarchical Plan Orchestration via D-Conductor and Q-Performer2024ROER: Regularized Optimal Experience Replay2024Energy-Guided Diffusion Sampling for Offline-to-Online Reinforcement Learning2024DIAR: Diffusion-model-guided Implicit Q-learning with Adaptive Revaluation2024Active Reinforcement Learning Strategies for Offline Policy Improvement2024