Dsrl: Offline Safe Reinforcement Learning Benchmarks for Robotics

Dsrl provides datasets and environment wrappers for offline safe reinforcement learning in robotics domains, authored by liuzuxin and updated in November 2025. The collection focuses on safety-constrained trajectories and imitation learning benchmarks for evaluating safe policy optimization.

Use Cases

• Training offline reinforcement learning models using safety-constrained trajectories
• Benchmarking imitation learning algorithms using robotics-specific state-action pairs
• Evaluating safe policy optimization using environment wrappers to measure constraint violations

Strengths

• Apache-2.0 license allows for open research and commercial use
• Includes environment wrappers for immediate integration with Safe RL benchmarks
• Focuses on safety-critical robotics domains, providing specialized data for constraint-based learning

Limitations

• Specific record counts and data dimensionality are not provided in the primary metadata
• Lack of column-level documentation requires manual inspection of the source code for schema identification

Provenance

Source

liuzuxin via GitHub

Freshness

Updated as recently as November 2025.