Moving beyond static math.
We are retiring the static Q&A format. alrospace-0.3 ships as time-series data (HDF5). Every row is a complete, multi-variable flight episode containing state vectors, sensor readings (IMU/GPS), and actuator commands. It is built from the ground up to be piped directly into Reinforcement Learning (RL) or Imitation Learning pipelines.
Dual 2D & 3D Trajectory Data
We are providing both formats to support your entire training lifecycle:
- 2D Dynamics: Perfect for rapid prototyping, debugging, and initial RL policy training where compute is at a premium.
- 3D 6DOF Dynamics: Full rigid-body dynamics for high-fidelity validation and real-world deployment.
NASA-Grade Physics Models
Simplified point-mass models are useless for real-world control. We are integrating true 6-Degrees-of-Freedom (6DOF) rigid body dynamics.
- Gravity: EGM2008 spherical harmonics (Degree 70+).
- Atmosphere: NRLMSISE-00 with real space weather (F10.7/Ap index).
- Ephemeris: SPICE/DE440 for Sun/Moon third-body perturbations.
The AI will train on the exact same physics models used by mission planners.
Expert GNC Policies
To train an autopilot, the AI needs an expert to mimic. We are generating action data using classical, proven flight algorithms. This includes PEG (Powered Explicit Guidance) for orbit insertion, G-FOLD for pinpoint powered landings, and EKF (Extended Kalman Filters) for navigation. The AI learns from the same guidance laws that flew the Space Shuttle.
Domain Randomization
A model that only flies in perfect weather is a failure. alrospace-0.3 includes full Monte Carlo randomization. Every episode injects wind shear (GRAM), sensor noise, thrust misalignment, and off-nominal failures (engine-out). Your model will learn to survive the edge cases.
Certified Against NASA Tools
We do not guess. Every generated trajectory is cross-validated against Basilisk (NASA's open-source 6DOF simulator) and GMAT. If our trajectory drifts by more than 1 meter from the reference simulator, the row is destroyed. The cryptographic Merkle proofs now seal the entire trajectory block, not just a single equation.
Associated Data Model
This article references the ALR Aerospace Platform 0.2. You can stream these verified reasoning traces immediately using our API.
[ FETCH DATA VIA API ]