Robot Hackathon: Edge of Intelligence

about this event

Run OpenVINO models on a real robot arm. Viam and Intel are hosting a two-day, hands-on hackathon for engineers who want to run vision and robot-policy models on real hardware. Every team gets a robot arm, a gripper, a depth camera, and a compute machine with an Intel® Core™ Ultra Series 3 processor and integrated Arc GPU and NPU. You fine-tune or train a model, export it to OpenVINO, and run it on the provided machine through Viam.

This is for robotics engineers, ML engineers, and software engineers. You do not need prior robotics hardware experience. We provide the hardware and a working starting pipeline, so you spend the event on the model and the task, not on drivers. Intel and Viam engineers will be on site as mentors and judges. Space is limited! Apply through this page. We accept applicants on a rolling basis until November 10, 2026.

Your hardware: Every team will have 2-4 people and receive the same kit from Viam: a robot arm, a gripper, a depth camera, and a compute machine with an Intel® Core™ Ultra Series 3 processor and integrated Arc GPU and NPU. Please bring your own laptop.

Tracks: Pick one of three tracks.

1. Teach it by showing. Record demonstrations of the packing task on your arm with a leader arm. Train a robot policy with ACT, or fine-tune SmolVLA. Export the policy to OpenVINO and run it through Viam on the Arc GPU. You start from a recorded dataset and two working baseline policies, and we score you on the tiers of the packing task you complete. The tiebreaker is fewest demonstrations.

2. Big models, small box. Run pi0.5, a 3-billion-parameter vision-language-action model, on the NUC's integrated GPU. Have it complete the packing task from spoken or typed instructions. Teams start from a checkpoint that Viam pre-fine-tuned on the hackathon kit. A named prize goes to the first team to run MolmoAct 2 on OpenVINO through Viam. This track is capped.

3. Spot it, sort it. Defect detection and response, with an industrial and lab focus. Build a pipeline that finds defective items with cameras and an anomaly detection model (Anomalib), decides what to do about each one with a vision-language or vision-language-action model, and has the arm sort good items into the box and defective ones into a reject bin. Every model runs on the NUC through OpenVINO. The defect scenario and items are revealed at the start of the event. Scored on detection accuracy, correct responses, latency, how you split work across NPU, GPU and CPU, and a short design explanation.

The packing task: Tracks 1 and 2 share one task: pack the order. A bin holds mixed items, an open shipping box sits next to it, and item positions are randomized. Each judged run is ten minutes. You earn points by tier: pick one item and place it in the box; pack the three items on the order and ignore two items that are not; fit an awkward item (a bottle laid flat, a bag, or a deformable item); close the lid; recover when a judge moves an item mid-run; follow a conditional spoken order, such as "pack everything except the fragile one" (track 2 only). A freestyle demo counts for part of your score. Show the judges something the tiers do not cover.

Build with OpenVINO: OpenVINO is Intel's inference toolkit. It runs trained models on Intel CPUs, Arc GPUs and NPUs. Viam handles the robot side: cameras, arms, grippers, configuration and deployment. Every track ends the same way. You export a model to OpenVINO, and Viam runs it on the NUC's Arc GPU and connects it to the camera and the arm. The starting pipeline we hand you already runs end to end, so your first job is swapping in your own model.

Prizes: Each member of the winning team in each track will receive a SO-101 robotic arm, a RealSense camera, and $5000 in Viam credits. Wildcard prizes, open to every team: Best reusable OpenVINO module published to the Viam registry during the event; Best hand-built "talk to your robot" pipeline — speech to text, vision-language model, detection and motion, all running locally.

Schedule — Friday, November 13: 9:00 AM check-in and breakfast; 9:30 AM kickoff covering tracks, the packing task, and a walkthrough of the starting pipeline; 10:00 AM teams and hardware; 10:30 AM hacking begins; 1:00 PM lunch; 6:00 PM dinner; 9:00 PM hard close — robots power down, doors close (no overnight hacking).

Saturday, November 14: 9:00 AM doors open, breakfast; 9:30 AM hacking resumes; 12:30 PM lunch; 3:30 PM demos and drinks — judging runs for all three tracks, freestyle demos; 5:00 PM awards; 6:00 PM event concludes.

What skills you need: No robotics experience required, but you should be comfortable with Python (the training, export and Viam SDK code you will write), Linux (SSH into the NUC and run terminal commands), and Git (basic workflows so your team can build together). Bonus: training or fine-tuning a vision model in PyTorch (not required). Bonus: prior OpenVINO or Viam experience (not required, and mentors can help you get started).

Before the event: We recommend taking our free online course Viam 101 (est. time: 45-90 mins) ahead of time. The first few lessons cover the configuration and SDK calls you will use on day one.

How to join: Space is limited. Apply through this page. We accept applicants on a rolling basis until November 10, 2026. In your application, tell us which track interests you and what you have built before. Attendees must be approved in advance to participate. We send confirmations to accepted participants.

Presented by Viam — the robotics platform that handles the robot side: cameras, arms, grippers, configuration and deployment — and Intel, maker of OpenVINO, the inference toolkit that runs trained models on Intel CPUs, Arc GPUs and NPUs.

tracks

when

tickets

prizes

faq