m1sc
Reach-down fixed-wing training tiles v2
A dataset of 2,500 synthetic terrain tiles used to train a reach-down fixed-wing landing-value Vision Transformer, comprising 500 synthetic terrains with five aircraft states per terrain. The dataset includes noisy observed terrain as inputs and Hamilton-Jacobi reach-avoid oracle labels as privileged targets.
Downloads30
Episodes2500
Why This Matters for Physical AI
This dataset provides synthetic training data for learning safe landing policies for fixed-wing aircraft by combining terrain perception with reach-avoid analysis, supporting research in autonomous aerial vehicle control and trajectory planning under uncertainty.
Technical Profile
- Modalities
- terrain_elevationterrain_slopeterrain_roughnesshazard_masksreach_margin
- Robot Embodiments
- fixed-wing_aircraft
- Environment
- simulation
- Task Types
- landingreach-avoidnavigation
- Episodes
- 2500
- Data Format
- npz
- Annotation Types
- reach_avoid_labelshazard_labelsmargin_labelsterrain_metadataaircraft_state
Access
Need custom terrain_elevation data?
Claru builds purpose-built datasets for simulation applications with dense human annotations and quality assurance.
Request a Sample Pack