A trajectory memory for robots: banks sensor-action sequences as vectors labeled by safety, then retrieves the safest neighbor to suggest the next move.
Vector search hackathons fill up with document retrieval, so embedding robot sensor and action sequences already moves Qdrant somewhere the judges had not seen five times that day. The stronger idea is folding a safety label into what retrieval optimises for: the nearest neighbour is easy, but the nearest neighbour that was also safe is what makes the result usable as a suggested next move.
Summaries are written by this site. Project pages include demo videos, screenshots and the team's own write-up (videos may autoplay).