Search papers, labs, and topics across Lattice.
This paper introduces Fisher Preserving Guidance with Outer Product Span Projection, a training-free inference method for diffusion models that constrains updates to the training manifold, addressing issues of unreliable trajectories in visual navigation tasks. The method approximates the Fisher-preserving update via a low-rank Jacobian factorization, enabling efficient computation with a single backward pass. Experiments across various navigation benchmarks demonstrate improved performance over existing diffusion policy baselines without requiring additional training.
Diffusion models can navigate more reliably without retraining, thanks to a clever guidance method that keeps them on the training manifold.
Diffusion models are effective for waypoint prediction in visual navigation, but standard sampling and test time guidance can produce unreliable or inefficient trajectories when updates drift off the training manifold. We propose Fisher Preserving Guidance with Outer Product Span Projection, a training-free inference method that avoids large Fisher drift associated with off-distribution actions while optimizing a task objective. Our method computes the Fisher-preserving update via a low-rank Jacobian factorization, requiring only a single backward pass per step and enabling real-time use. We further introduce Truncated Fisher Denoising Sensitivity as an uncertainty signal and use it for robust multi-sample action blending. Experiments on toy and realistic navigation benchmarks, including Maze2D with TSDF-based guidance, PushT with official Diffusion Policy weights, and visual navigation in simulation and on real robots, demonstrate consistent improvements in performance over strong diffusion-policy baselines without additional training.
