Robot policy performance can be improved by up to 60% by identifying a single 'golden ticket' constant noise vector instead of sampling from a Gaussian.
arXiv · March 18, 2026 · 2603.15757
The Takeaway
It reveals that frozen, pretrained diffusion/VLA policies contain high-performance behaviors that are hidden by standard sampling; finding one constant vector improves results without any retraining or fine-tuning.
From the abstract
What happens when a pretrained generative robot policy is provided a constant initial noise as input, rather than repeatedly sampling it from a Gaussian? We demonstrate that the performance of a pretrained, frozen diffusion or flow matching policy can be improved with respect to a downstream reward by swapping the sampling of initial noise from the prior distribution (typically isotropic Gaussian) with a well-chosen, constant initial noise input -- a golden ticket. We propose a search method to