Brachiosaurus Model
A sauropod-inspired quadrupedal simulation model designed for four-legged locomotion and head-positioning tasks.
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 | Quadrupedal |
|---|---|
| Observation dimension | 86 |
| Action dimension / actuators | 30 |
| Generalized coordinates (nq) | 38 |
| Generalized velocities (nv) | 37 |
| Compiled dynamic model mass | 175.3 kg |
| Plant contract revisions | Policy r8 · physics r4 · visual r2 |
| Model | environments/brachiosaurus/assets/brachiosaurus.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 — Balance | stand (deliverable) | — | Learn to stand on four legs without falling | 6M | reward ≥ 1,040; episode length ≥ 950; ≥ 10 episodes/evaluation; 3 consecutive passes |
| 2 — Locomotion | walk (deliverable) | 1 — Balance | Learn coordinated quadrupedal walking | 16M | reward ≥ 100; episode length ≥ 750; avg. velocity ≥ 0.75 m/s; ≥ 10 episodes/evaluation; 3 consecutive passes |
| 3 — Food Reach | hunt (deliverable) | 2 — Locomotion | Move the head tip within the configured distance threshold of food | 12M | reward ≥ 100; task success ≥ 50.0%; ≥ 10 episodes/evaluation; 3 consecutive passes |
Backend-specific task-success semantics
- Stable-Baselines3 / JAX/MJX — Head-tip distance-threshold success: The head-tip site comes within the configured food-reach threshold of the food target while the food-reach bonus is enabled.
Per-deliverable success semantics
- stand (1 — Balance) · Stable-Baselines3 — Reward-gated stance (reward_and_length/v1): The stance checkpoint clears the reward_and_length/v1 gate: mean evaluation reward at or above the statue-derived collapse rail and a near-full-horizon mean episode length over the required consecutive evaluations. The zero-action statue clears this gate, so stand is labelled by its gate kind here rather than claimed as certified stance quality; stance_quality/v1 waits on shin instrumentation (a kneeling pose currently reads identically to airborne) (plan §4.8).
- 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 — Food Reach) · Stable-Baselines3 / JAX/MJX — Head-tip distance-threshold success: The head-tip site comes within the configured food-reach threshold of the food target while the food-reach bonus 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 · 2026-07-18
Provenance: Historical model · unverified · 30 evaluation episodes · Stable-Baselines3 (version not recorded)
| Stage | Trained steps | Best eval reward | Avg. forward velocity | Task success (Stable-Baselines3) | Passed |
|---|---|---|---|---|---|
| 1 — balance | 6M | 1740.53 | 0.01 m/s | — | passed retired gate (reward gate) |
| 2 — locomotion | 16M | 6634.60 | 1.42 m/s | 3.3% | passed retired gate (reward gate) |
| 3 — food reach | 12M | 1368.13 | 0.71 m/s | 100.0% | passed retired gate (reward gate) |
Current Stable-Baselines3 definition for this task label: The head-tip site comes within the configured food-reach threshold of the food target while the food-reach bonus is enabled.
Features
- First quadrupedal species in the project
- Articulated 4-segment neck with head control (6 actuators)
- Front legs longer than rear, producing the model's raised-front posture
- Simplified column-like legs for support
- Head-positioning behavior using neck articulation
- Three published behaviors on one trunk (
configs/brachiosaurus/stages.toml): stand (stance), walk (locomotion, warm-started from stance) and hunt (behavior— the food-reach task, warm-started from locomotion); see Behavior Recipes
Unique Characteristics
Brachiosaurus differs from the bipedal species in the project:
- Quadrupedal gait instead of bipedal
- Target-proximity task using the head tip instead of a predatory contact task
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 | Torso height stability (m) |
success_rate | Fraction of episodes where the head tip enters the configured distance threshold around the food target |
min_prey_distance | Closest approach to food (m, head_food_distance) |
Training Videos
Published videos for Brachiosaurus Altithorax, where available. Artifact provenance is shown on each stage.
Balance
Learn to stand on four legs 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 coordinated quadrupedal walking
PPO · Stable-Baselines3 artifact · historical model · unverified · backend version not recorded. It is not evidence for the current model or stage config.
Food Reach
Move the head tip within the configured distance threshold of food
No published video is available for this stage.
Usage
cd environments/brachiosaurus
# View the model
python scripts/view_model.py
# Train the stance node using its current TOML-configured budget
python scripts/train_sb3.py train --stage 1
# The advancing stages (stance, locomotion, behavior) in manifest order
python scripts/train_sb3.py curriculum --algorithm ppo
# Reuse an earlier run's certified stance and walk; train hunt in a fresh directory
python scripts/train_sb3.py curriculum --algorithm ppo \
--trunk-from logs/<earlier_run> --output-dir logs/<new_run>
See Behavior Recipes for the notebook workflow.