Frequently Asked Questions
Common questions about robot learning, data collection, and our services.
What is robot learning?
Robot learning is the process of training robots to perform tasks from data (imitation learning) or trial-and-error (reinforcement learning). Key approaches include behavior cloning, DAgger, diffusion policies, and vision-language-action (VLA) models. See our Glossary for definitions.
How much data do I need for imitation learning?
It depends on task complexity. Simple pick-place: 50–200 demonstrations. Complex multi-step tasks: 200–1000+. Pre-trained models (OpenVLA, Octo) can often achieve good results with fewer demos via fine-tuning. We can help with data collection for your specific task.
Can I use simulation data for real robots?
Yes, but with caveats. Sim-to-real transfer works best with domain randomization and often benefits from real-world fine-tuning. Real data is often more reliable for deployment. See Why Real-World Data Beats Simulation Alone and our Sim-to-Real glossary.
What robot do I need for data collection?
WidowX, Franka, ALOHA, and similar arms are common. See our Robot Platforms Comparison for cost, DOF, and data-collection suitability. We provide data collection services if you prefer not to operate hardware yourself.
What is a VLA model?
Vision-Language-Action (VLA) models take images and language instructions as input and output robot actions. Examples: OpenVLA, RT-2, Octo. See our Models catalog and VLA & VLM glossary.
How do I get started with OpenVLA or Octo?
Both have Hugging Face checkpoints and GitHub repos. Check our OpenVLA and Octo pages for official links. Fine-tuning typically requires 50–500 demos for your robot and task.
Do you offer custom data collection?
Yes. We provide data collection services for imitation learning and RL. We can collect demonstrations on your hardware or ours, with learning-ready formatting and QA.
Where is Silicon Valley Robotics Center located?
Our flagship center is at 654 High St, Palo Alto, CA 94301 — near Stanford University. Same-day pickup for OpenArm, Mobile ALOHA, and more. We also serve Boston, Chicago, and Houston. See Locations.
Can I rent a robot for research?
Yes. We offer robot leasing — short-term and long-term. Unitree Go2, Booster K1, G1 humanoid. Try before you buy, or use for pilots and demos.
What is Open X-Embodiment?
Open X-Embodiment is a large-scale robot dataset combining RT-X, BridgeData, DROID, and others. Used to train foundation models like OpenVLA and Octo. See our Open X-Embodiment page and Models catalog.
Best robot for imitation learning?
WidowX, ALOHA, Franka, and OpenArm are common. Choice depends on cost, DOF, and task. See Robot Platforms Comparison and our Getting Started guide.