Minari: Standardized Offline Reinforcement Learning Data for Gymnasium

Farama-Foundation's Minari provides a standardized format and collection of reference datasets for offline reinforcement learning as of January 2026. It hosts trajectories from Gymnasium environments to support training agents without active environment interaction.

Use Cases

• Training offline RL agents using state and action trajectories
• Benchmarking algorithms using reward sequences and termination flags
• Developing imitation learning models from expert-level observation data

Strengths

• Standardized format for Gymnasium
• Maintained by Farama-Foundation
• Updated as of January 2026

Limitations

• Specific column names and environment-specific details are not listed in the metadata
• Requires the Minari library for optimal data loading

Provenance

Source

Farama-Foundation

Collection Method

synthetic

Freshness

Updated as of January 2026.