RoboHacks, Y Combinator
Winner of the overall prize and the Scale AI data track. Cross-robot spatial memory.
Built at RoboHacks, the robotics hackathon hosted at Y Combinator, where the project won both the Overall Prize and the Scale AI Data Track ($2,500 and two Innate robots).
The idea: a robot should not have to explore a room to know what is in it. We take an iPhone video walkthrough with spoken narration and turn it into a shared 3D map that any robot can use for pick and place tasks, so spatial memory transfers between robots instead of being rebuilt from scratch.
The pipeline combines NVIDIA Parakeet ASR for the narration, 3D reconstruction and ICP registration for the map, a Gemma 3 VLM for grounding language to objects in the scene, and an action chunking policy to execute the resulting tasks.