"VectorForge Robotics team on AI Infra Summit Hackathon Hackathon"

Team Idea

Running bimanual VLA manipulation policies at the edge without datacenter GPUs. We train a language-conditioned vision-language-action policy on scripted bimanual demonstrations, convert and quantize it with OpenVINO, and benchmark latency and task success on Intel silicon against the PyTorch baseline. Voice-driven task instruction via Speechmatics real-time transcription. Target: Intel Bimanual VLA Manipulation (online track) + Speechmatics bonus.

Team Members

Sadi Shihab

sadi-vector

Submission

Bimanual VLA Table-Setting: Measured, Not Assumed

Dual SO-101 bimanual table-setting in MuJoCo. Every design choice, from gripper pads to handover geometry to OpenVINO precision, came from a measurement, not an assumption. A trained, language conditioned ACT policy with an OpenVINO CPU/GPU/NPU benchmark.