Datasets:
Embodied Physics Trajectories — Protocol Company
Real physics data for autonomous AI. Generated by G^G (genesis_core), a pure Rust physics engine built from first principles across thirteen substrates of physical reality. No external physics libraries. Every equation traces to a published reference. Every trajectory is SHA-256 sealed from initial conditions to terminal state and independently verifiable.
The industry has been training autonomous AI on simulation built from gaming middleware designed for visual plausibility. These platforms are non-deterministic: run the same inputs twice, get different outputs. The data in this repository is different. Same seed → same trajectory → same SHA-256 hash, on any machine, every time.
Key Findings from the Corpus
Mars EDL ignition boundary: Standard approach produces 0.7% survival. First-principles integration finds the ignition cliff: 87.27% survival at the exact moment the boundary equation resolves.
Orbital attitude recovery: Simplified single-axis models predict ~100% tumble recovery. Full 3D Euler coupling with gyroscopic cross-coupling produces 17.6% at ω=0.5 rad/s.
Marine GPS-denied navigation: 0.1% of GPS-denied missions fail catastrophically (>500m unrecoverable error), invisibly, without warning, with a successful mission logged.
Atheric swarm collapse: 82.4% of multi-robot communication networks collapse from microseconds of clock drift.
Full findings: No Way Back: Crossing the Physics Gap in Physical AI — 49-page technical thesis, every finding SHA-256 anchored.
Thirteen Substrates
| Substrate | Physics | Format |
|---|---|---|
| Reactor kinetics | Neutron kinetics, thermal-hydraulics, SCRAM boundary | Failure envelope |
| Tokamak plasma | MHD stability, Grad-Shafranov, Z-axis shear | Failure envelope |
| Grid swing dynamics | Swing equation, inertia floor, renewable intermittency | Failure envelope |
| Orbital mechanics | Quaternion attitude, Euler coupling, J2–J4 | Trajectory corpus |
| Drone swarm dynamics | EKF spoofing, Friis topology, GPS denial | Failure envelope |
| Humanoid robotics | ZMP stability, 3D Euler gait, yaw coupling | Trajectory corpus |
| Marine GPS-denied nav | 6-DOF hydrodynamics, IMU drift accumulation | Trajectory corpus |
| Electromagnetic signal | Shannon capacity, Friis path loss, frequency hopping | Trajectory corpus |
| Soil contact mechanics | Boussinesq stress, glomalin coupling, 4 soil types | Trajectory corpus |
| Mars EDL | CO₂ atmospheric drag, 1000 Hz Euler, ignition boundary | Trajectory corpus |
| Mycelial network | Kirchhoff hyphal flow, percolation threshold | Trajectory corpus |
| Deep-time celestial | Yoshida 4th-order symplectic, 1PN corrections | Trajectory corpus |
| Asteroid N-body | O(N²) pairwise gravity, Metal GPU authoritative | Trajectory corpus |
Verification
Every trajectory in this dataset carries a SHA-256 proof hash computed at generation time. The hash is a fingerprint of every timestep from initial conditions to terminal state. Any modification to any timestep changes the hash.
On-chain anchor: canister ad7wi-4aaaa-aaaad-aeijq-cai on the Internet Computer. Immutable. Publicly queryable. No infrastructure dependency on Protocol Company.
Products and Licensing
Custom physics generation — $750,000+/engagement
Edge case libraries — $250,000+/substrate
Simulation and training data audit — $75,000–$300,000/year
- Downloads last month
- 26