GEAR-SONIC

Real CPU planner_sonic.onnx run inside the MuJoCo-backed G1 showcase, driven by an explicit staged velocity-tracking script instead of a single constant command. This page keeps the full charts, playback, and downloadable artifacts.

Generated: 2026-04-24 02:27:15Z Commit: f1b940ac0f78 Actions run

GEAR-SONIC

NVIDIA GR00T · Planner-only velocity tracking on the published G1 planner contract

gear_sonic

BAD REAL MUJOCO SIM VELOCITY_SCHEDULE runtime.velocity_schedule

Real CPU planner_sonic.onnx run inside the MuJoCo-backed G1 showcase, driven by an explicit staged velocity-tracking script instead of a single constant command.

Robotunitree_g1_gear_sonic
Ticks400
Avg inference4.060 ms
Achieved rate44.69 Hz
Target positions
Body vx command vs actual
Yaw rate command vs actual
Inference latency
Embedded Rerun viewer Fetches run.rrd lazily when the viewer enters the viewport.
Preparing interactive view Loads the viewer runtime and recording on demand when visible.
Case key gear_sonic
Policy family gear_sonic
Command kind velocity_schedule
Expected behavior Planner-only velocity tracking on the published G1 planner contract
Showcase transport MuJoCo sim
Embodiment assets/robots/unitree_g1/g1_29dof.xml
MuJoCo model variant meshless-public-mjcf+ground-plane
Command data [2.000, 0.000, 0.000, 0.000, 0.600, 0.000, ...]
Checkpoint source Published GEAR-SONIC ONNX checkpoints
Demo family Velocity tracking
Demo sequence Stand, accelerate from 0.0 to 0.6 m/s over 2 s, command a 90 degree right turn over 1 s, accelerate into a 1.0 m/s run over 3 s, then settle back to stand.
Model artifact models/gear-sonic/planner_sonic.onnx
Command source runtime.velocity_schedule
First target frame [-0.169, -0.085, 0.223, 0.880, -0.157, -0.089, ...]
Last target frame [-0.203, 0.061, 0.092, 0.146, 0.206, -0.029, ...]
Quality verdict BAD
Verdict basis achieved rate 44.7 Hz < 45
Mean joint error 0.119 rad
Joint error p95 0.414 rad
Peak joint error 2.769 rad
Min base height 0.713 m
Frames height < 0.4 m 0
VX RMSE 0.306 m/s
Yaw RMSE 0.772 rad/s
Heading change 2.5 deg
Forward distance 3.087 m

Visual checkpoints

Each image overlays the target pose in blue against the actual replayed pose in orange. The checkpoints are selected from the replay trace so you can cross-check startup, motion onset, peak latency, furthest progress, and final state without opening a second report.

start_tick_0000

initial state before meaningful motion

tick 0 0.00 s
start_tick_0000 track overlay
track
start_tick_0000 side overlay
side
start_tick_0000 top overlay
top

peak_latency_tick_0002

highest inference latency (109.511 ms)

tick 2 0.04 s
peak_latency_tick_0002 track overlay
track
peak_latency_tick_0002 side overlay
side
peak_latency_tick_0002 top overlay
top

first_motion_tick_0003

first mean joint delta >= 0.08 rad (0.084 rad)

tick 3 0.06 s
first_motion_tick_0003 track overlay
track
first_motion_tick_0003 side overlay
side
first_motion_tick_0003 top overlay
top

fallback_mid_25_tick_0100

deterministic fallback at 25% of the replay because the evidence checkpoints collapsed to the same frame

tick 100 2.00 s
fallback_mid_25_tick_0100 track overlay
track
fallback_mid_25_tick_0100 side overlay
side
fallback_mid_25_tick_0100 top overlay
top

furthest_progress_tick_0399

largest planar displacement from start (3.042 m)

tick 399 7.98 s
furthest_progress_tick_0399 track overlay
track
furthest_progress_tick_0399 side overlay
side
furthest_progress_tick_0399 top overlay
top