Delta Flow β Humanoid Benchmark
Delta Flow (ImageWAM / HumanWAM whole-body delta-flow policy) checkpoints trained on Humanoid Benchmark tasks (40 epochs; the last two saved checkpoints are published).
Checkpoints
100 demonstrations per task. Every row is one training run; the epoch folders are its published checkpoints.
| Folder | Task | Robot | Cameras (model input) | Global batch | Proprio augmentation | Checkpoints |
|---|---|---|---|---|---|---|
b6-pedal-bin/ |
B6 β pick up a chip barrel, step on the pedal bin, drop it in | Unitree G1 (Inspire hands) | ego_view, chest | β | aligned (GR00T A20 recipe) | epoch-030, epoch-040 |
b7-black-clothes-hamper/b256/ |
B7 β take the black garment off the hanger and drop it into the hamper (re-train, current protocol) | Unitree G1 (Inspire hands) | ego_view, chest | β | aligned (GR00T A20 recipe) | step-006000, step-008000 |
a10-trolley-push-v3/ |
A10 v3 β push a platform trolley (148 episodes) | Unitree G1 (Inspire hands) | ego_view, chest | β | aligned (GR00T A20 recipe) | step-006000, step-008000 |
a3-fridge-open-v4/b256/ |
A3 v4 β open the fridge door with the right hand (125 episodes) (re-train, current protocol) | Unitree G1 (Inspire hands) | ego_view, chest | β | aligned (GR00T A20 recipe) | step-006000, step-008000 |
a7-cabinet-close-v4/b256/ |
A7 v4 β close the open cabinet door with the right hand (re-train, current protocol) | Unitree G1 (Inspire hands) | ego_view, chest | β | aligned (GR00T A20 recipe) | step-006000, step-008000 |
A bNNN/ sub-folder (NNN = its global batch, e.g. b256/, b512/) is a re-train of that task under the current benchmark protocol (2026-11-12: global
batch GR00T / pi0.5 / Delta Flow / ACT 256, Psi-Zero 512, DiT4DiT 64; proprioception augmentation aligned with the
GR00T A20 recipe) and is the one to use for that task; the checkpoints directly under the task folder are the earlier
training round, kept for reference.
What an epoch folder contains
Every checkpoint folder (epoch-XXX/, or step-XXXXXX/ for runs trained for a fixed number of steps) is self-contained for evaluation and deployment: weights, model configuration, normalization
statistics and protocol.json (task, data, input/output dimensions, cameras, batch, steps, augmentation, file list).
Simple training records are included for reproducibility (training plan, training log, configs, metrics); optimizer /
scheduler / RNG / DeepSpeed resume state is not. code/ holds the adapter code used for training and inference
(code/VERSIONS.json: upstream commits and package versions).
Inputs and augmentation
Model inputs are identical across all benchmark models: G1 state 44-D (29 body joint angles + 6 + 6 measured hand values
- 3 projected gravity), H2 state 78-D (75 joint angles + 3 projected gravity), no joint velocities. Cameras: G1 ego view +
chest; H2 head + chest (the 3-camera re-trains of 2026-11-12/13 also used the right-wrist camera; the
protocol.jsonof each folder lists its cameras). The third-person camera recorded in the datasets is for data review only and is never a model input. Each camera is a separate image input, except for Delta Flow, DiT4DiT and FastWAM, which place the views side by side in one frame. Training uses the Humanoid Benchmark online data augmentation spec v2 (15 photometric / geometric / sensor / lighting effects, one of them boosted to 1.5x per sample) with the model's own image augmentation turned off, plus joint-angle bias and noise on the proprioception (column "Proprio augmentation"; the earlier recipe also delayed the proprioception and dropped action-history rows).
legacy/ holds earlier experiments that do NOT follow the benchmark protocol; see legacy/README.md.