Datasets:

spiderpilot89
/

gg-physical-ground-truth

language:
  - en
pretty_name: G^G Physical Ground Truth (Teaser Pack)
tags:
  - physics-simulation
  - trajectory-data
  - monte-carlo-simulation
  - autonomous-systems
  - robotics
  - reinforcement-learning
  - humanoid-robotics
  - autonomous-vehicles
  - drones
  - aerospace
  - orbital-mechanics
  - hypersonics
  - sim-to-real
  - physical-ai
  - sha256-verified
  - rust
  - apple-silicon
task_categories:
  - robotics
  - reinforcement-learning
  - time-series-forecasting
license: mit
size_categories:
  - 10M<n<100M
configs:
  - config_name: humanoid
    data_files:
      - split: train
        path: data/humanoid_centroidal_joint_dynamics.parquet
  - config_name: vehicle
    data_files:
      - split: train
        path: data/autonomous_vehicles_heavy_logistics.parquet
  - config_name: drone
    data_files:
      - split: train
        path: data/drones_adversarial_contested_dynamics.parquet
  - config_name: orbital
    data_files:
      - split: train
        path: data/orbital_systems_deep_space.parquet
  - config_name: hypersonic
    data_files:
      - split: train
        path: data/hypersonic_aerothermal_ablation.parquet
  - config_name: reactor
    data_files:
      - split: train
        path: data/reactor_isotopic_failure_envelope.parquet
  - config_name: tokamak
    data_files:
      - split: train
        path: data/tokamak_shear_failure_envelope.parquet
  - config_name: swing
    data_files:
      - split: train
        path: data/swing_intermittency_failure_envelope.parquet
homepage: https://www.zerotrustphysics.com

G^G Physical Ground Truth (Teaser Pack)

First-principles physical ground truth trajectory datasets for autonomous systems and Sim-to-Real calibration. Generated in pure Rust, sealed with SHA-256 hash chains, and verified on-chain.

Training robot brains in visual simulation environments (like MuJoCo or NVIDIA Isaac Sim) leads to the Sim-to-Real Cliff when physical limits are uncalibrated. When your policy transitions from standard simulator environments to real hardware, it encounters non-deterministic slippage, thermal degradation, control loop jitter, sensor latency, and mechanical wear that simplified dynamics models can ignore.

G^G provides sovereign physical ground truth datasets modeling raw physics failures at 1000Hz (1ms step latency) to close the Sim-to-Real gap.

This teaser pack contains approx. 30,067,000 trajectories (approx. 4.7 GB total) across eight critical domains.

Quick Start

You can load any of the eight configurations directly using the Hugging Face datasets library:

from datasets import load_dataset

# 1. Load Humanoid Centroidal & Joint Dynamics (50D)
humanoid = load_dataset("johnkruze/gg-physical-ground-truth", "humanoid", split="train")
print(humanoid[0])  # View first joint state step

# 2. Load Autonomous Vehicles & Heavy Logistics (51D)
vehicles = load_dataset("johnkruze/gg-physical-ground-truth", "vehicle", split="train")

# 3. Load Tactical Drones & Contested Airspace (59D)
drones = load_dataset("johnkruze/gg-physical-ground-truth", "drone", split="train")

# 4. Load Orbital Systems & Deep Space (20D)
orbital = load_dataset("johnkruze/gg-physical-ground-truth", "orbital", split="train")

# 5. Load Hypersonic Aerothermal Ablation (47D)
hypersonic = load_dataset("johnkruze/gg-physical-ground-truth", "hypersonic", split="train")

# 6. Load Fission Reactor Isotopic Degradation (5D)
reactor = load_dataset("johnkruze/gg-physical-ground-truth", "reactor", split="train")

# 7. Load Tokamak Fusion Z-Shear & Quench (5D)
tokamak = load_dataset("johnkruze/gg-physical-ground-truth", "tokamak", split="train")

# 8. Load Swing Grid Intermittency & Cascade (4D)
swing = load_dataset("johnkruze/gg-physical-ground-truth", "swing", split="train")

Dataset Configurations

1. Humanoid Centroidal & Joint Dynamics — 50D

Binary: humanoid_impedance_monte_carlo
Volume: 15,000 trajectories (approx. 305 MB)
Physics: Sagittal plane (xz) centroidal dynamics, pelvic pitch quaternion tracking, and Whole-Body Control (WBC) active joint-impedance. Simulates slip, recovery, and pelvic buckle/thermal failure on variable-stiffness terrain. Models gear backlash wear, ankle tendon elongation/creep, liquid payload slosh moments, and battery cell thermal runaway. Ground friction coefficient is swept ($\mu \in [0.10,0.60]$).

2. Autonomous Vehicles & Heavy Logistics — 51D

Binary: vehicle_hydroplane_monte_carlo
Volume: 10,000 trajectories (approx. 1.2 GB)
Physics: 4-wheel chassis 6-DOF dynamics, wheel rotational states, dynamic normal loads, road moisture, and EKF estimator divergence signals. Tire interaction uses Pacejka Magic Formula force-slip equations ($F_x, F_y$ at all four corners). Models speed-dependent aquaplaning friction decay, hydraulic stiction during breakaway, and thermal damper fade (viscosity collapse above 120°C). Autopilot EKF is blind to lateral slip.

3. Tactical Drones & Contested Airspace — 59D

Binary: drone_canopy_monte_carlo
Volume: 20,000 trajectories (approx. 587 MB)
Physics: Euler-Boussinesq wind-shear tensor coupling, ground cushion aerodynamics, dynamic dust obscuration, and motor thermal limits. Payload weights swept continuously (1.5 kg to 35.0 kg) with Parallel Axis Theorem inertia tensor updates during payload release (CoG migration). Active cyber-physical attacks: RF jamming (EW) inflating packet drop rates and control latency (up to 195ms), and adversarial GPS spoofing bias vectors triggering EKF divergence. Includes crystal oscillator clock drift (Bechmann polynomial).

4. Orbital Systems & Deep Space — 20D

Binary: orbital_tumble_monte_carlo
Volume: 12,000 trajectories (approx. 788 MB)
Physics: Spacecraft attitude gyroscopic coupling integration (Euler equations) under fuel depletion, cross-inertia, solar storm torque, and fuel slosh dynamics. Models fuel slosh as a mass-spring-damper oscillator coupled to spacecraft attitude equations.

5. Hypersonic Aerothermal Ablation — 47D

Binary: hgv_plasma_monte_carlo
Volume: 10,000 trajectories (approx. 484 MB)
Physics: Reentry 6-DOF trajectory propagation under Sutton-Graves convective heating, CoG migration from asymmetric ablation, nose cone shell thinning, wing flutter bend/twist, and EKF covariance trace tracking under plasma blackout GPS denial.

6. Fission Reactor Isotopic Degradation — 5D

Binary: reactor_monte_carlo
Volume: 10,000,000 trajectories (approx. 156 MB)
Physics: 6-group delayed precursor point-kinetics coupled with fuel ($T_f$) and coolant ($T_c$) temperatures. Simulates negative Doppler and coolant feedbacks, Iodine-135/Xenon-135 neutron poisoning chains, and prompt criticality runaways under malicious or incorrect control rod extraction.

7. Tokamak Fusion Z-Shear & Quench — 5D

Binary: tokamak_monte_carlo
Volume: 10,000,000 trajectories (approx. 438 MB)
Physics: Radial MHD force balance, unstable vertical elongated displacement ($z$), active PF coil PD stabilizer loops, and resistive superconducting coil heating ($I^2R$). Models coil thermal quench (transition to $1.0\ \Omega$ at 15.0 K) and subsequent confinement wall breach due to edge controller precision limits (quantization noise).

8. Dynamic Swing Grid Phase Cascade — 4D

Binary: swing_monte_carlo
Volume: 10,000,000 trajectories (approx. 712 MB)
Physics: Rotor angle swing equation, Phase-Locked Loop (PLL) tracking error, steam turbine governor droop feedback, and variable inverter-based resource (IBR) penetration. Sweeps rotational inertia degradation down to zero ($H=0$ on inverters), modeling the fast phase collapse horizon where AI routing latency cannot arrest desynchronization.

Verification & Trust

Every trajectory in this dataset carries a SHA-256 proof_hash. The proof chain works as follows:

The G^G physics engine generates state vectors deterministically from seed parameters.
The final state vector is hashed: SHA-256(trajectory_data).
The hash is chained sequentially: SHA-256(prev_proof + current_proof).
The master proof chain anchor is stored immutably on the Internet Computer (ICP).

Backend Canister: ad7wi-4aaaa-aaaad-aeijq-cai
SOMA Ledger: mwtw4-wiaaa-aaaak-qx57a-cai (ICRC-1/2/3)

Licensing

This dataset is released under the MIT License. You are free to download, inspect, modify, and use this data for both research and commercial purposes.

See the LICENSE file for the full terms.

Citation

@dataset{kruze_gg_physical_ground_truth_2026,
  author = {Kruze, John},
  title = {G^G Physical Ground Truth: Trajectory Datasets for Embodied AI and Sim-to-Real Calibration},
  year = {2026},
  url = {https://huggingface.co/datasets/johnkruze/gg-physical-ground-truth},
  note = {First-principles Rust physics, SHA-256 proof chains, ICP verification}
}

About John Kruze

John Kruze builds physics ground truth for autonomous systems. 13 physics domains. First-principles Rust. SHA-256 sealed at every integration step. No game engines, no approximations.

The core thesis: autonomous systems operating without human-in-the-loop require deep physical reasoning, not visual game engine approximations. When GPS is denied, when communication is jammed, when the wind shear hits — your autopilot must reason from physics priors alone.

Website: www.zerotrustphysics.com
GitHub: github.com/johnkruze/genesis-core (engine source) · github.com/johnkruze/gg-physics (framework docs)
Author: John Kruze (LinkedIn)