Denoising Particle Filters: Learning State Estimation with Single-Step Objectives
Lennart Röstel, Berthold Bäuml
AI summary
Problem
Learning-based state estimation typically relies on expensive end-to-end sequence training that lacks modularity and interpretability, while training models on single-step transitions often causes predictions to drift away from the data manifold.
Approach
The method trains dynamics and measurement models independently using single-step objectives and combines them at inference through a diffusion-based particle filtering process that corrects particle predictions toward the measurement data manifold.
Key results
- Competitive or superior accuracy to end-to-end baselines on challenging robotic tasks
- Efficient, scalable training using only single-step objectives without sequence unrolling
- Flexible integration of known sensor models and priors without retraining
- Mitigation of distributional drift via likelihood-constrained diffusion and biased prior guidance
Why it matters
Provides robotics researchers with a modular, interpretable, and computationally efficient alternative to end-to-end state estimation that seamlessly integrates classical filtering properties with modern learning.
Abstract
Learning-based methods commonly treat state estimation in robotics as a sequence modeling problem. While this paradigm can be effective at maximizing end-to-end per- formance, models are often difficult to interpret and expen- sive to train, since training requires unrolling sequences of predictions in time. As an alternative to end-to-end trained state estimation, we propose a novel particle filtering algorithm in which models are trained from individual state transitions, fully exploiting the Markov property in robotic systems. In this framework, measurement models are learned implicitly by minimizing a denoising score matching objective. At inference, the learned denoiser is used alongside a (learned) dynamics model to approximately solve the Bayesian filtering equation at each time step, effectively guiding predicted states toward the data manifold informed by measurements. We evaluate the proposed method on challenging robotic state estimation tasks in simulation, demonstrating competitive performance compared to tuned end-to-end trained baselines. Importantly, our method offers the desirable composability of classical filtering algorithms, allowing prior information and external sensor models to be incorporated without retraining.