Parkour-STL: Quadruped Robot Trajectories with Formal Specifications

Parkour-STL is a dataset of Unitree Go2 quadruped parkour rollouts generated in IsaacSim 6.0 and IsaacLab 3.0. The dataset pairs these trajectories with a library of normalized, STL-monitorable predicates sampled at 10 Hz. It was created by seil-umd and last updated in May 2026.

Use Cases

• Training RL agents for parkour tasks based on simulated quadruped trajectories.
• Validating robot behavior against formal specifications using the STL-monitorable predicates.
• Benchmarking formal verification tools for robotic control systems.

Strengths

• Trajectory data is paired with formal specifications (STL predicates).
• Predicates are normalized and sampled at a consistent 10 Hz frequency.
• Data is generated using a specific simulation environment (IsaacSim 6.0 / IsaacLab 3.0).

Limitations

• Dataset card is a v0.1 draft and may be incomplete.
• Column-level documentation is absent; field semantics must be inferred after download.
• Row count and file formats are unknown, which may limit suitability assessment.

Provenance

Source

seil-umd

Collection Method

Rollouts generated in IsaacSim 6.0 / IsaacLab 3.0 simulation.

Freshness

Last updated 2026-05-29 11:11:21