This paper teaches AI teams to get better by scoring every move they make, not just the final answer.
RoboTracer is a vision-language model that turns tricky, word-only instructions into safe, step-by-step 3D paths (spatial traces) robots can follow.