Velociraptor Model
A dinosaur-inspired bipedal simulation model with actuated sickle-claw geoms, trained for locomotion and target-contact tasks. Its manifest (configs/velociraptor_mongoliensis/anatomical/stages.toml) declares three published behaviors on one trunk: stance (stand), locomotion (walk, warm-started from stance) and behavior (hunt — the strike task, warm-started from locomotion). See Behavior Recipes.
The stance node certifies stance quality, not reward: since physics r3 its gate is stance_quality/v2, which judges
every episode of a 40-episode panel on the floor's force under each foot (no hop at reset, both feet planted, both
toes and the metatarsal head loaded, no splay or drift, no actuator held at its limit) and certifies the stance when
at least 37 of 40 are clean. Velociraptor trains on Stable-Baselines3 only.
Generated specifications
These values are generated from the environment and compiled MJCF model. “Dynamic mass” is a simulator property, not an anatomical weight estimate.
| Gait | Bipedal |
|---|---|
| Observation dimension | 70 |
| Action dimension / actuators | 22 |
| Generalized coordinates (nq) | 31 |
| Generalized velocities (nv) | 30 |
| Compiled dynamic model mass | 13.5 kg |
| Plant contract revisions | Policy r12 · physics r4 · visual r4 |
| Model | environments/velociraptor_mongoliensis/anatomical/assets/velociraptor_mongoliensis_anatomical.xml |
Current curriculum configuration
The Stable-Baselines3 budgets and early-advancement gates below come from the current TOML files and can differ from older published runs. A stage also ends when its configured budget is exhausted. The recipe column names the behavior a stage belongs to and whether its checkpoint is a published deliverable; the warm-start column names the stage it starts from.
| Stage | Recipe | Warm-start from | Objective | SB3 configured budget | SB3 early-advancement gate |
|---|---|---|---|---|---|
| 1 — Stance | stand (deliverable) | — | Learn to stand and balance without falling | 12M | clean stance episodes LCB95 ≥ 0.8 over ≥ 40 episodes (settle 100 steps); an episode is clean when it reaches the horizon with min_all_feet_support ≥ 0.98, max_touchdown_rate ≤ 0.25, max_window_displacement_m ≤ 0.1, min_foot_load_share ≥ 0.4, max_actuator_saturation_fraction ≤ 0.1, max_settle_airborne_substeps ≤ 0, max_settle_peak_floor_force_bw ≤ 2, min_foot_load_share_windowed ≥ 0.35, max_foot_contact_fraction ≤ 0.02, max_phantom_support_fraction ≤ 0.05, max_nonfoot_load_fraction ≤ 0.01, max_settle_stance_width_change_m ≤ 0.05, min_support_geom_duty ≥ 0.5, min_support_geom_coverage ≥ 0.8, min_foot_load_share_statue_ratio ≥ 0.8; full-horizon episodes ≥ 95.0%; reward rail ≥ 1,710; reward rail ≥ 0.6 × the statue's; verdict from the floor-truth stance_gate_report.json on the handoff pair (post-stage; fail-closed when absent) |
| 2 — Locomotion | walk (deliverable) | 1 — Stance | Learn forward walking/running | 8M | reward ≥ 100; episode length ≥ 750; avg. velocity ≥ 2 m/s; ≥ 10 episodes/evaluation; 3 consecutive passes |
| 3 — Strike | hunt (deliverable) | 2 — Locomotion | Sprint and strike prey with sickle claw | 12M | reward ≥ 100; task success ≥ 50.0%; ≥ 10 episodes/evaluation; 3 consecutive passes |
Backend-specific task-success semantics
- Stable-Baselines3 — Sickle-claw contact success: A left or right sickle-claw geom contacts the prey geom while the strike reward is enabled.
Per-deliverable success semantics
- stand (1 — Stance) · Stable-Baselines3 — Stance quality (stance_quality/v2): The stance checkpoint clears the stance_quality/v2 gate, judged post-stage on the handoff pair: the one-sided 95% lower bound on the share of the 40 certification-panel episodes that are clean on floor truth (horizon reached, no hop or impact in the settle window, both feet down and loaded, both toes and the metatarsal head of each foot loaded, no actuator held at its limit) meets the configured bound, with the reward rails as collapse floors only. The clean count and its bound are recorded in gate_verdict.json but not exported to summary.json until a later phase (decision D-A9, deferred by D-B15), so the catalog names them with no value.
- walk (2 — Locomotion) · Stable-Baselines3 — Gated forward velocity (reward_and_length/v1): The locomotion checkpoint clears the reward_and_length/v1 gate: mean forward velocity at or above the stage's configured minimum, with its reward and episode-length floors, over the required consecutive evaluations.
- hunt (3 — Strike) · Stable-Baselines3 — Sickle-claw contact success: A left or right sickle-claw geom contacts the prey geom while the strike reward is enabled.
Published run summaries
These are historical experiment records. A result is not evidence for the current model revision unless its provenance is explicitly marked current and verified.
PPO · Stable-Baselines3 · 2026-03-15
Provenance: Historical model · unverified · evaluation episode count not recorded · Stable-Baselines3 (version not recorded)
| Stage | Trained steps | Best eval reward | Avg. forward velocity | Task success (Stable-Baselines3) | Passed |
|---|---|---|---|---|---|
| 1 — Stance | 6M | 1964.43 | 0.11 m/s | — | passed retired gate (reward gate) |
| 2 — locomotion | 8M | 2678.68 | 3.47 m/s | — | passed retired gate (reward gate) |
| 3 — strike | 8M | 1366.19 | 2.02 m/s | 93.3% | passed retired gate (reward gate) |
Current Stable-Baselines3 definition for this task label: A left or right sickle-claw geom contacts the prey geom while the strike reward is enabled.
SAC · Stable-Baselines3 · 2026-03-21
Provenance: Historical model · unverified · evaluation episode count not recorded · Stable-Baselines3 (version not recorded)
| Stage | Trained steps | Best eval reward | Avg. forward velocity | Task success (Stable-Baselines3) | Passed |
|---|---|---|---|---|---|
| 1 — Stance | 6M | 970.19 | -0.64 m/s | — | passed retired gate (reward gate) |
| 2 — locomotion | 8M | 2078.62 | 2.91 m/s | — | passed retired gate (reward gate) |
| 3 — strike | 8M | 1195.43 | 1.63 m/s | 90.0% | passed retired gate (reward gate) |
Current Stable-Baselines3 definition for this task label: A left or right sickle-claw geom contacts the prey geom while the strike reward is enabled.
Anatomy
- Torso - Elongated body with simplified head/neck
- Legs - Digitigrade hind limbs (hip pitch/roll, knee, ankle, 2 toe digits per leg); since physics r3 the leg springs are anchored at the standing pose and the toes stand flat, so the leg servos hold the stance
- Feet - Touch sensors on both toes and the metatarsus of each foot, summed into one contact reading per foot; an elliptic friction cone since physics r4, the one friction model of every species
- Sickle claws - Actuated claw geom on digit 2 of each foot
- Arms - Stub forelimbs with shoulder pitch/roll actuators
- Tail - 5-segment counterbalance (4 actuated segments)
Diagnostic Metrics (behavior node, hunt)
After each training stage the eval command reports:
| Metric | Description |
|---|---|
mean_forward_velocity | Average forward speed (m/s) |
gait_symmetry | Left/right stride symmetry ∈ [0, 1] |
stride_frequency | Step frequency (Hz) |
cost_of_transport | Energy efficiency (lower is better) |
mean_pelvis_height | Upright stability (m) |
mean_heading_alignment | cos θ toward prey ∈ [-1, 1] |
success_rate | Fraction of episodes with successful sickle-claw target contact |
min_prey_distance | Closest approach to prey (m) |
Training Videos
Published videos for Velociraptor Mongoliensis (Anatomical), where available. Artifact provenance is shown on each stage.
Stance
Learn to stand and balance without falling
PPO · Stable-Baselines3 artifact · historical model · unverified · backend version not recorded. It is not evidence for the current model or stage config.
Locomotion
Learn forward walking/running
PPO · Stable-Baselines3 artifact · historical model · unverified · backend version not recorded. It is not evidence for the current model or stage config.
Strike
Sprint and strike prey with sickle claw
PPO · Stable-Baselines3 artifact · historical model · unverified · backend version not recorded. It is not evidence for the current model or stage config.
Usage
# From the repository root. View the model
python environments/velociraptor_mongoliensis/scripts/view_model.py --embodiment anatomical
# Train the stance node using its current TOML-configured budget
python -m environments.shared.train --species velociraptor_mongoliensis --embodiment anatomical train --stage 1
# The advancing stages (stance, locomotion, behavior) in manifest order
python -m environments.shared.train --species velociraptor_mongoliensis --embodiment anatomical curriculum --algorithm ppo
# Reuse an earlier run's certified stance and walk; train hunt in a fresh directory
python -m environments.shared.train --species velociraptor_mongoliensis --embodiment anatomical curriculum --algorithm ppo \
--trunk-from logs/<earlier_run> --output-dir logs/<new_run>
See Behavior Recipes for the notebook workflow.