Now in private beta

Deploy to
robot arms
instantly

Push your code to GitHub. We deploy it to real SO-101 robot arms, run evaluations, and send back video feeds & sensor data.

Like Vercel, but for physical robots.

No hardware needed
LeRobot compatible
daslab-deploy

How it works

From code to robot evaluation in minutes. No hardware setup, no complex configurations.

1

Connect your repository

Link your GitHub repo containing your robot policy or control code. We support LeRobot, PyTorch, and standard Python packages.

daslab connect github.com/you/robot-policy
2

Push your code

Every push triggers automatic deployment to our robot arm cells. We handle environment setup, dependency installation, and weight loading.

git push origin main
3

Watch it run

Real-time video feeds from multiple angles. Monitor your policy's performance as it executes on physical hardware.

4

Get results

Receive comprehensive evaluation data: success rates, trajectory logs, sensor readings, and video recordings. All synced back to your repo.

# Results in daslab-results/
├── videos/
├── trajectories.json
├── metrics.json
└── sensor_data.h5
Updated weekly

VLA Benchmarks

Standardized evaluations of vision-language-action models on real hardware. All models tested under identical conditions on our SO-101 cells.

SO-101 Standard Tasks50 trials per task
Last updated: Dec 2024
#ModelOrganizationPick CubeStack BlocksPour WaterOverallTrend
👑OpenVLA-7BStanford96.2%89.4%82.1%89.2%
2RT-2-XGoogle DeepMind94.8%91.2%78.5%88.2%
3Octo-BaseBerkeley AI92.1%85.7%79.3%85.7%
4RoboFlamingoTsinghua89.5%82.3%75.8%82.5%
5GR-1Fourier87.2%78.9%73.2%79.8%

Want your model evaluated? Submit it for benchmarking.

Built for researchers

Everything you need to iterate faster on robotic manipulation research.

Instant deployment

Your code runs on hardware within minutes of pushing. No waiting for lab access or hardware setup.

Multi-angle video

4K video feeds from multiple camera angles. Perfect for debugging and generating training data.

Reproducible benchmarks

Consistent evaluation conditions across all runs. Compare models fairly with standardized metrics.

Rich sensor data

Joint positions, forces, gripper state, and more. All synchronized and exportable in standard formats.

LeRobot native

First-class support for LeRobot policies. Just point us to your checkpoint and we handle the rest.

Custom scenes

Coming soon: Configure custom task scenes, object arrangements, and evaluation criteria.

Ready to deploy?

Join the waitlist for early access. We're onboarding research teams weekly.

Academic teams get priority access. Include your lab affiliation.

SOC 2 compliant
Your data, your IP
<5 min deploy time