Muninn: Your Trajectory Diffusion Model But Faster

2026-05-11

AI Summary

Diffusion-based trajectory planning struggles to meet real-time robotic control demands due to the computational overhead of its iterative denoising process. This work proposes a training-free dynamic caching mechanism that, during sampling, constructs an uncertainty budget based on internal trajectory representation changes within the diffusion model and error propagation coefficients to dynamically determine whether to reuse previously computed denoising results. The approach establishes, for the first time, a training-agnostic caching strategy coupled with a provable upper bound on uncertainty, effectively balancing generation quality and computational efficiency. Evaluated across multiple benchmarks, the method achieves up to a 4.6× speedup without compromising task performance or safety, and demonstrates robust effectiveness in real-world robot navigation and manipulation tasks.

Abstract

Diffusion-based trajectory planners can synthesize rich, multimodal robot motions, but their iterative denoising makes online planning and control prohibitively slow. Existing accelerations either modify the sampler or compress the network--sacrificing plan quality or requiring retraining without accounting for downstream control risk. We address the problem of making diffusion-based trajectory planners fast enough for real-time robot use without retraining the model or sacrificing trajectory quality, and in a way that works across diverse state-space diffusion architectures. Our key insight is that diffusion trajectory planners expose two signals we can exploit: a cheap probe of how their internal trajectory representation changes across steps, and analytic coefficients that describe how denoiser errors affect the sampler's state update. By calibrating the first signal against the second on offline runs, we obtain a per-step score that upper-bounds how far the final trajectory can deviate when we reuse a cached denoiser output, and we treat this bound as an uncertainty budget that we can spend over the denoising process. Building on this insight, we present Muninn, a training-free caching wrapper that tracks this uncertainty budget during sampling and, at each diffusion step, chooses between reusing a cached denoiser output when the predicted deviation is small and recomputing the denoiser when it is not. Across standard benchmarks Muninn delivers up to 4.6x wall-clock speedups across several trajectory diffusion models by reducing denoiser evaluations, while preserving task performance and safety metrics. Muninn further certifies that cached rollouts remain within a specified distance of their full-compute counterparts, and we validate these gains in real-time closed-loop navigation and manipulation hardware deployments. Project page: https://github.com/gokulp01/Muninn.

Problem

• diffusion-based trajectory planning
• real-time robot control
• trajectory quality preservation
• model acceleration
• state-space diffusion architectures

Innovation

• trajectory diffusion
• real-time planning
• caching acceleration
• uncertainty budget
• training-free speedup