d4rl

🌟 Align diffusion processes with detailed human preferences to improve machine learning models for richer, more accurate outputs.

A collection of robotics simulation environments for reinforcement learning

Split conformal prediction for off-policy evaluation in offline RL (D4RL MuJoCo). Stage 1 of 4: episode reconstruction and data foundation, with guards for two silent D4RL failure modes.

Robot RL portfolio: bipedal locomotion, algorithm comparison, offline RL, gait phase estimation (4 reproducible projects)

Diagnostic study of regime representations in offline reinforcement learning under heterogeneous behavior data.

[IcETRAN 2026] Official implementation of Flow Matching Policy for Behavioral Cloning paper.

Clean single-file implementation of offline RL algorithms in JAX

ANA 699 Capstone — Offline reinforcement learning for robotics using Decision Transformers and MuJoCo

Unified offline RL / offline-to-online RL course project scaffold with D4RL smoke tests

Offline RL via sequence modeling: BC, Decision Transformer, and Online DT comparison on D4RL benchmarks

Online goal-reaching RL with diffusion planning: a 2D point-mass prototype plus a Maze2D Diffuser workflow driven by an autonomous agentic experiment controller.

D4RL benchmark but ported to work end to end with gymnasium

[NeurIPS 2025] A human-like RL framework that improves human-likeness while achieving strong performance, and can be easily integrated into various RL algorithms

Unified Implementations of Offline Reinforcement Learning Algorithms

Diffusion-Guided Tree Search: Uncertainty-Aware Planning with Learned World Models

Velocity-parameterized diffusion for trajectory planning, with MPPI sampling and periodic replanning. Matches or beats Diffuser (ICML 2022) on Maze2D at half the diffusion steps.

a clear and fast jax/flax version of [Diffusion-Policies-for-Offline-RL](https://github.com/Zhendong-Wang/Diffusion-Policies-for-Offline-RL)

Codes accompanying the paper "Score Regularized Policy Optimization through Diffusion Behavior" (ICLR 2024).

Learning from Sparse Offline Datasets via Conservative Density Estimation (ICLR 2024)

High-quality single-file implementations of SOTA Offline and Offline-to-Online RL algorithms: AWAC, BC, CQL, DT, EDAC, IQL, SAC-N, TD3+BC, LB-SAC, SPOT, Cal-QL, ReBRAC