protocolcompany/embodied-physics-trajectories

Embodied Physics Trajectories

100+ million physics trajectories across 10 substrates, generated from 200KB of hand-rolled Rust on a single Apple M4 Pro.

No external physics libraries. No game engines. No Unity. No MuJoCo. Pure Euler and symplectic integration, hand-written in Rust, compiled to native Metal compute. Every trajectory is SHA-256 sealed at generation time.

This is not synthetic data from a rendering pipeline. This is physics ground truth — bodies in substrates experiencing forces.

What This Dataset Contains

Time-series trajectory data from Monte Carlo physics simulations across 10 physical domains. Each trajectory records the full state evolution of a body moving through a physically-grounded environment, with timestep-level sensor readings, phase transitions, anomaly flags, and cryptographic proof hashes.

Substrate	Physics	Bodies	Integration	Rate
Mars (CO₂)	Atmospheric drag, supersonic deceleration, suicide burn	EDL Lander	Euler 1000Hz	144/s
Orbital (Vacuum)	Quaternion attitude, reaction wheels, gyroscopic coupling	MAVEN-class satellite	Euler 100Hz	353/s
Marine (Seawater)	Pressure gradients, buoyancy, IMU drift, GPS-denied nav	Submarine, Autonomous Boat	Euler 1000Hz	11/s
Terran (Mycelium)	Boussinesq soil stress, glomalin coupling, moisture transport	8 body types (tracked, wheeled, legged, drone, humanoid, agrover, scout drone, titanhauler)	100Hz	10,750/s
Atheric (Signal)	Shannon capacity, Friis path loss, frequency hopping	Coherent Signal	—	15,830/s
Mycelial (Network)	Hyphal conductance, Kirchhoff flow, percolation threshold	Mycorrhizal Mesh	—	760/s
Asteroid (NEO)	O(N²) N-body gravity, Hookean contact mechanics	Rubble Pile	Metal GPU	9/s
Celestial (Astro)	Yoshida 4th-order symplectic, N-body ephemeris	Spacecraft	—	71/s
Plutonian (Core)	Entropy decay, phase crystallization, deep-time evolution	Phase System	—	71/s
Energy (Grid)	Swing equation, nuclear/solar/wind/hydro dispatch	Grid Node	—	186,060/s

Generation rates are live-measured on Apple M4 Pro (14-core CPU, 20-core GPU, 24GB unified memory).

Schema

Every exported JSON file follows this structure:

{
  "dataset_metadata": {
    "generator": "G^G {Substrate} Monte Carlo v1.0",
    "domain": "{substrate}",
    "scenario": "{scenario_type}",
    "trajectories": 10000,
    "physics_engine": "genesis_core::{module} ({integrator} {rate}Hz)",
    "version": "1.0.0",
    "generated_at": "2026-03-07T23:24:59Z"
  },
  "trajectories": [
    {
      "id": "{substrate}_{hash}",
      "type": "{trajectory_type}",
      "scenario": "{scenario_name}",
      "steps": 306171,
      "score": { ... },
      "proof_hash": "0ec728887ad144603b56ee53e302b72342c7fc81...",
      "reasoning_context": {
        "anomaly_type": "HARD_LANDING",
        "is_anomaly": true,
        "snapshot": { ... }
      },
      "data": [
        { "t": 0.0, "alt": 24371.8, "vel": 180.6, "phase": "ENTRY", ... },
        { "t": 1.0, "alt": 24191.2, "vel": 181.3, "phase": "ENTRY", ... }
      ]
    }
  ]
}

Per-Substrate Timestep Fields

Mars EDL — Entry, Descent, and Landing through CO₂ atmosphere:

t, alt, vel, pos[3], rho (atmospheric density), phase (ENTRY/CHUTE/RETRO/TERMINAL)

Orbital — Satellite tumble recovery in vacuum:

t, omega[3] (angular velocity), omega_mag, wheel_h[3] (reaction wheel momentum),
fuel_kg, battery_pct, in_sunlight, measured_mag, phase (TUMBLE/DETUMBLE/FINE_POINT)

Marine — GPS-denied underwater/surface navigation:

t, depth, v (velocity), p (position), drift, nav_error, current[3],
battery, waypoint (index)

Terran — Robot-soil contact mechanics:

t, pressure_pa, yield_pa, compaction, moisture, phase (APPROACH/CONTACT/TRAVERSE)

Score Fields (Per-Substrate)

Substrate	Score Fields
Mars	dispersion, final_velocity, max_deceleration_g, mission_success
Orbital	final_omega_rad_s, fuel_remaining_pct, recovered, recovery_time_s
Marine	nav_error_m, waypoints_reached, waypoints_total, energy_consumed_pct, mission_success
Terran	pressure_yield_ratio, compaction, seed_emerged, mission_success

Reasoning Context

Every trajectory includes a reasoning_context block — a structured snapshot designed for training autonomous decision-making systems. It captures the anomaly state at a critical moment:

{
  "anomaly_type": "HARD_LANDING",
  "is_anomaly": true,
  "snapshot": {
    "alt": 24371.9,
    "chute_deployed": true,
    "dust_storm": false,
    "retro_fired": false,
    "vel": 302.5
  }
}

This enables training models that reason through embodied physics rather than about it — the distinction between information retrieval and sovereign decision-making.

Scale

Milestone	Trajectories	Status
100M Corpus (proof-only)	100,009,600	Complete — SHA-256 sealed, data not exported
Exported Corpus	~85,200	Complete — 10 substrates, ~10.6 GB
Mars 10M Product Run	10,000,000	Complete — 60 GB raw JSON
Terran 10M Product Run	10,000,000	Complete — 24 GB raw, 1.4 GB compressed
Marine 10M Product Run	—	Pending (estimated ~4.2 TB)

Master Proof Hash (100M Sequential): 4b12bf0f6392cd1a0e76f5cffcdfc93868a1a83aa509e620df4743182ca39b30

Use Cases

• Reinforcement learning — Pre-training world models with physically-grounded trajectories
• Sim-to-real transfer — Training autonomous systems on diverse physics before deployment
• Anomaly detection — Learning failure modes from labeled anomalous trajectories
• Autonomous decision-making — Dark Window scenarios (GPS-denied, comm-loss) where systems must reason from physics priors alone
• Robotics foundation models — Multi-substrate pre-training across land, sea, air, and space
• Spacecraft autonomy — MAVEN-class tumble recovery, orbital rendezvous
• Agricultural robotics — Soil-robot interaction across 4 soil types × 8 body configurations
• Marine autonomy — GPS-denied submarine and surface vessel navigation

Technical Details

Engine: G^G Genesis Core — pure Rust, no external physics dependencies Machine: Apple M4 Pro (14-core CPU, 20-core GPU, 24 GB unified memory) GPU: Metal compute shader for asteroid N-body (O(N²) gravity, GPU-authoritative) Proofs: SHA-256 chain — every trajectory is cryptographically sealed at generation time On-chain anchor: 42 autonomous decision manifests verified on Internet Computer (IC) mainnet Canister: ad7wi-4aaaa-aaaad-aeijq-cai (IC mainnet, stable storage)

Sample Data

The samples/ directory contains small extracts (5 trajectories each, 10 timesteps per trajectory) from each substrate for schema validation and preview.

Licensing

This dataset is commercially licensed. Sample data is provided for evaluation.

For full dataset access, contact: kruze@aijesusbro.com

Pricing:

Product	Price
Mars EDL (10M trajectories)	$5,000
Terran Soil (10M trajectories, 8 body types × 4 soils)	$5,000
Marine Navigation (10M trajectories)	$10,000
Full Corpus	Contact for licensing

Website: aijesusbro.com

Citation

@dataset{protocolcompany_embodied_physics_2026,
  author = {Kruze, John Frances IV},
  title = {Embodied Physics Trajectories: Monte Carlo Simulation Data for Autonomous Systems},
  year = {2026},
  publisher = {Protocol Company},
  url = {https://huggingface.co/datasets/protocolcompany/embodied-physics-trajectories}
}

About Protocol Company

Protocol Company builds embodied reasoning infrastructure — physics engines that generate the training data autonomous systems need to operate in the real world. 10 physics substrates, 100+ million trajectories, zero external dependencies. The simulation is the product.