A unified benchmarking and evaluation framework for VLA policies across simulation and real-world robots.
HEALTH ? 69 / Stable
// momentum
// why it's here
- Active: last commit today
- Health 69/100 with only 7 stars
- 5 forks for 7 stars: people build on it
// how to help
// readme
FluxThemis
FluxThemis runs simulation and real-environment evaluation for FluxVLA. The model boundary is a versioned ZMQ protocol: FluxThemis owns environments and episode scheduling, while FluxVLA owns model loading, preprocessing, inference, action denormalization, and authoritative benchmark artifacts.
Responsibilities
| Component | Owns |
|---|---|
| FluxVLA | Model config, checkpoint, inference replicas, preprocessing, statistics, action units, authoritative evaluation journal and benchmark summary |
| FluxThemis | Evaluation config, simulator assets and layouts, environment processes, episode scheduling, action execution, videos, progress, resource monitoring and resume |
Every prediction request contains the complete observation and prompt. With the default dynamic scheduling, consecutive requests from one episode may run on different model workers. Fixed scheduling is also available when supported and configured on both sides.
Installation
For development, install only the core package and ZMQ runtime:
conda create -n fluxthemis-dev python=3.10 -y
conda activate fluxthemis-dev
cd /root/projects/FluxThemis
python -m pip install -e .