AgibotTech / sonic_for_a3

# SONIC adaption for A3
**035 / 200k · A3-fast RKNN · Real-robot teleoperation** — Watch the video (3:32).
Explore the 024 / 035 training curves — searchable, interactive TensorBoard scalar snapshots.
> **Note — Real-robot safety:** We strongly recommend using a safety gantry / fall-arrest harness and appropriate protective measures during real-robot teleoperation. This checkpoint does **not** guarantee stable tracking of every motion; unexpected behavior, loss of balance, and falls can occur. **Before playing any motion or deploying your own teleoperation integration on the real robot, first validate it in MuJoCo.** MuJoCo simulation cannot fully represent real-robot behavior, and successful simulation does not guarantee safe hardware execution. Keep physical safety protection in place, maintain a safe distance from the robot, and prioritize the personal safety of operators and bystanders.
This project builds on
GR00T-WholeBodyControl / SONIC.
We sincerely thank the GEAR team for their amazing and solid work, their
comprehensive open-source release, and their thorough documentation, which
made this A3 adaptation possible.
This repository is a focused public release of the SONIC whole-body-control
workflow for **AgiBot A3**. It is intentionally scoped to the 035 training contract, its A3 simulation assets, MuJoCo sim2sim,
and the matching step-200,000 deploy artifacts.
Start with the end-to-end guide. It provides
the authoritative path from A3 flat CSV data to training, sim2sim, ONNX/RKNN,
and a staged deploy package.
The identifier `035` is an experiment label only; it has no special meaning
and does not denote a product version, model generation, or hardware variant.
## What is included
### 1. A3 training code adaptation
The release contains the complete A3 adaptation of the SONIC training workflow:
Hydra configurations, portable launchers, passive-foot URDF/MJCF assets,
selected flat CSV examples, the 035 checkpoint/config location, and the
A3-specific reward and observation contracts. It provides download commands for the pretrained
035 step-200,000 checkpoint trained on the internal motion dataset, targeting
basic walking and simple manipulation abilities; the PT is hosted on
Hugging Face.
Although this checkpoint was trained using the A3 URDF, its domain randomization
was specifically tuned to support both **A3 and A3 Ultra (A3U)**. The corresponding
models can therefore also run on A3 Ultra.
### 2. MuJoCo sim2sim
It includes MuJoCo sim2sim support with the closed-loop ankle/waist serial and
parallel joint solver, so the same policy interface can be exercised offline
against the A3 model before deployment.
### 3. High-performance C++ deployment on A3
The C++ deployment framework is designed for high-performance inference on A3.
It supports cross-compilation for the A3 RK3588S Rockchip board, including NPU
inference through RKNN, and a Drive Thor target for A3 Ultra through the Thor
runtime variant. **The Rockchip / RKNN deployment workflow