YAML Metadata Warning:empty or missing yaml metadata in repo card
Check out the documentation for more information.
XBrain-v1 model resources
This package provides the model resources for the PiperX (piper_x), Piper
(piper), and ARX X5 (arx_x5) adapters in XPolicyLab.
Upload the contents of this directory to the model repository root. The
package contains piper_x/model_ema/, piper/model_ema/, arx_x5/model_ema/,
norm/, tokenizer/, fast_tokenizer/, and a resource checksum manifest.
The inference runtime and policy adapter are supplied by the XBrain-v1
XPolicyLab submission; model weights alone are not executable policy code.
After downloading the repository to a local directory:
python3 /path/to/release/verify_release.py
source /path/to/release/xbrain_env.sh
verify_release.py validates the SHA256 and size of every payload file. A
quick metadata-only check is available with --fast; that mode skips files
larger than 1 GiB and is not a substitute for complete download verification.
xbrain_env.sh resolves paths from its own location and exports the matching
checkpoint and normalization file for each robot, plus the shared tokenizer
paths. Launch the existing XPolicyLab policy server in that same shell using
the desired env_cfg_type. It does not start a server or control a robot.
Each checkpoint produces 50 actions. The adapter selects the outgoing prefix and applies task-specific gripper processing according to its policy code. Observations use the standard 14-dimensional dual-arm joint state. The policy returns absolute joint targets and absolute gripper values; do not add the current joint state again in the executor.
The checkpoint inference_config.json paths are relative to this package
root. The official adapter receives resolved absolute paths from the exported
environment variables and does not need the original development filesystem.
For a separate direct pipeline, resolve these relative paths against the
package root before passing them to the pipeline constructor.
This package has its own release_manifest.json. The previous release's
download_checkpoint.sh pins a different manifest and normalization file
layout. Download this package using the model repository's ordinary download
method, run the verification above, and source the included environment file.
For a private repository, use an account with access when downloading.
- Downloads last month
- -