EVA Client

EVA-Client: A Unified Framework for Deployment, Evaluation, and Data Collection on Real Robots

Paper GitHub
eva console · live demo
0:00 / 0:00

Why we built this

One Client, Full Cycle.

Training frameworks have converged: OpenPI, LeRobot, StarVLA, and VLA Foundry solve much of the training-side stack. The real-robot side should not still be a patchwork of project-specific scripts. EVA-Client fills that missing infrastructure for embodied-policy iteration: collect teleop data, inspect and prepare datasets, deploy checkpoints, compensate latency, smooth trajectories, run model evaluations, compare logs, and feed results back into the next training round. One client covers the full real-robot iteration cycle.

eval logs → next round
1Bring your robot
AgileX ARX Franka UR5e Galaxea AgiBot
teleop
2DataEVA
Data CollectionTeleop capture & quality check Episode ReplayInspect & relabel

prepares training-ready data

training set
3Policy training
OpenPI
starVLA
GR00T
DreamZero
checkpoint
4DeploymentEVA
sync strategy
EVA
async strategies
run on real hardware
5Real-robot evaluationEVA
AGILEX
ARX
Franka
UR5e
Galaxea
AgiBot

One closed loop: your robot feeds data, EVA organizes training-ready datasets, policies train, EVA smooths and deploys checkpoints on real hardware, and evaluation logs flow back to start the next round — real-robot infrastructure for the full model-iteration cycle.

↓ Get started View source

Citation

If EVA-Client is useful for your research or product, please cite:

@misc{evaclient2026,
  title        = {EVA-Client: A Unified Framework for Deployment, Evaluation,
                  and Data Collection on Real Robots},
  author       = {EVA-Client Contributors},
  year         = {2026},
  howpublished = {\url{https://github.com/Noietch/EVA-CLIENT}},
}