10 Reference: model, parameters and licensing
10.1 Reference model files
The simulation uses MuJoCo MJCF (.xml + .stl meshes), sharing the hardware's own design:
| File | Description |
|---|---|
hardware/v1/robot_v1.xml | Main model (geometry/mass matched to the hardware) |
hardware/v1/meshes/*.stl | 6 STL meshes that assemble into 4 printed groups (base plate/shell, shoulder mount, upper-arm bracket/elbow mount, forearm) |
hardware/v1/meshes/print_mass_model.json | Per-part printed mass table (15% infill) |
No URDF is bundled; for ROS use, export one from the CAD/meshes above.
10.2 Firmware parameters (firmware/sr_support_b_ptk/include/config.h)
| Parameter | Value | Meaning |
|---|---|---|
POLICY_DT / CTRL_PERIOD_US | 0.02 s / 20000 µs | 50 Hz control period |
MAX_CMD_RATE | 0.08 rad/tick | Output slew limit (4 rad/s) |
FALL_UP_Z | 0.30 | Fall-gate up_z threshold |
FALL_W_QUIET | 0.9 rad/s | Fall-gate angular-rate threshold |
FALL_CONFIRM_TICKS | 25 | Fall-gate confirmation (0.5 s) |
EXEC_STEPS_PER_WP | 20 | Ticks per reference waypoint |
EXEC_RETRACT_RATE | 0.033 rad/tick | Retract rate |
EXEC_DONE_TICKS | 50 | Consecutive standing ticks to finish |
EXEC_TIMEOUT_MS | 14000 | Per-reference timeout |
EXEC_COOLDOWN_TICKS | 75 | Cooldown after done/timeout (1.5 s) |
EXEC_SWITCH_TICKS | 15 | Pose-switch confirmation ticks |
EXEC_SWITCH_UP | 0.60 | Max up_z for a switch |
EXEC_SWITCH_W | 0.8 rad/s | Max angular rate for a switch |
EXEC_MIN_SWITCH_AGE | 25 | Earliest tick for a switch |
EXEC_MAX_SWITCHES | 5 | Max switches per run |
EXEC_NEUTRAL_MAX | 120 | Max neutral-retract ticks |
SERVO_SHOULDER_CENTER_US | 1681 | Shoulder zero pulse (visual-center value baked 2026-11; re-zero per page 05) |
SERVO_CENTER_US | 1456 | Elbow zero pulse (same) |
SERVO_US_PER_RAD | 636.62 | 2000 µs / π |
Q_SHOULDER_MIN/MAX | −1.85 / +1.82 rad | Shoulder manual-command limits |
Q_ELBOW_MIN/MAX | −0.90 / +0.70 rad | Manual range (executor uses ±1.55 mechanically) |
SERVO_SHOULDER_SIGN/ELBOW_SIGN | −1 / −1 | Servo direction |
IMU_YAW90 | 1 | IMU mounted 90° rotated |
MAHONY_KP / ACCEL_TRUST_BAND | 6.0 / 2.5 m/s² | Attitude filter: accel-correction gain / trust band |
WIFI_AP_SSID / WIFI_AP_PASS | SR-Support / srsupport | Hotspot (w toggles it; on at boot) |
10.3 Reference sequences (shoulder, elbow; rad; value-for-value with training/b_line/getup_reference_b39.py)
| Pose | Sequence |
|---|---|
roll_pos | [1.75,-0.90] → [-0.25,-0.90] → retract |
roll_neg | [-1.80,0.70] full-speed direct press (hold ≤40 ticks) → wedge pumps [-1.80,0.35] → [-1.80,0.05] ×2 (each hold ≤30) → fallback: [0.35,0.70] → [0.35,1.10] → [-1.80,0.05] (whip-slam, hold ≤40) → final hold ≤90 → retract |
pitch_pos | [-1.32,1.45] → [-0.25,1.45] → retract |
pitch_neg | [-1.32,-1.65] → [-0.25,-1.65] → retract |
10.4 Verified status and factory inspection (2026-11-06)
- Factory inspection: each finished unit gets - servo zeros defined and baked per unit (page 05), gyro + accel zero calibration (NVS), and a four-direction get-up test before shipping;
- All four directions verified on the real robot, each as a direct get-up (no roll-over), about 1–6 s; no servo envelope violations;
roll_negfinal recipe: full-travel direct press + elbow-wedge pumps ×2, whip-slam retry as fallback (the earlier roll-over + re-plan scheme is retired);- The attitude pipeline was corrected with accel-bias calibration (see page 06, 6.2); lying poses now match the real robot;
roll_negstill does not reproduce in the 1× mass simulation model (arm mass/friction differ); that is a known model gap - the hardware firmware is validated on hardware.
10.5 Licensing
- B-line firmware/training code: MIT / Apache-2.0;
- B-line CAD/STL: CERN-OHL-S-2.0 (strong reciprocity — check obligations before commercial use);
- SelfRisingRobot: no license; this tutorial contains none of its code, waypoints or models;
- Patent CN109693729A/B: perform an FTO review before commercial use.