Ola Rønning

Wave-Robust Passive AUV Localization Using FP-MUSIC

2026-09-23T00:00:00+00:00

Localizing an autonomous underwater vehicle without pre-deployed seabed transponders, or direct access to onboard vehicle sensors remains a core challenge. We present a receiver-passive 3-D localization and spatial mapping system utilizing a single floating surface buoy equipped with a hydrophone array and an inertial measurement unit (IMU). The central difficulty is that surface wave motion induces six-degree-of-freedom (6-DOF) perturbations that rotate the array between snapshots, degrading conventional subspace processing. We resolve this by introducing a fixed-point iterative MUltiple SIgnal Classification algorithm (FP-MUSIC) that uses IMU measurements to de-warp snapshot covariances prior to direction-of-arrival estimation. Furthermore, we employ a subspace-projected wideband matched filter to resolve beacon ranges and use power asymmetry for independent front-back identification. Evaluations across simulated sea states demonstrate that FP-MUSIC substantially reduces localization error relative to uncompensated methods and sustains robust 3-D tracking and vehicle orientation estimation under wave-induced motion. At moderate sea state, FP-MUSIC increases the 2-m beacon-separation accuracy from approximately 45% to 75%.

You Should Be Properly Scoring Your Odometry

2026-09-22T00:00:00+00:00

When we evaluate the performance of our odometry, it is common practice to score the estimated track against a ground truth. Unfortunately, scoring uses point metrics, such as the root mean square error, that ignore the covariance matrix which estimators like filters and smoothers already report. Using the covariance matters for two reasons. First, the covariance encodes the estimator's uncertainty, so it tells us whether the estimator trusts its own output. An overconfident estimator will not report itself lost. Second, the covariance weights the error in each direction of the estimate. Without the covariance, an estimator is unduly penalized for a high error in an uncertain direction. Instead of point metrics, we should use strictly proper scoring rules. These rules score the estimate together with its reported uncertainty. Strictly proper scoring rules recover the point metrics when no covariance is reported, and they diagnose covariance inconsistency when covariance is reported. Using a one-sided pairwise test, we show that two estimators can expose overconfidence in at least one of them without a ground truth. Strictly proper scoring rules and our pairwise test are available in our open-source framework smfeval. As a case study, we use smfeval to assess the uncertainty quality of the translational component of ground-based LiDAR-inertial odometry. Across four filters we find overconfidence - the worst case reports centimeter certainty with kilometer error. Knowing the filters are overconfident, we investigate the mechanism. The investigation traces overconfidence to filters crediting LiDAR measurements with more new information than they carry.

Pose Estimation Beyond Deep Learning

2026-03-01T00:00:00+00:00

Underwater robots have no reliable GPS, and visual localization breaks down in turbid, low-light water. Sonar is the natural fallback, though acoustic sensing brings its own difficulties, from low signal-to-noise to elevation ambiguity, that make good measurement models hard to build. Purely data-driven pose regression is one route, but it depends on large labeled datasets that are scarce underwater.

This project develops probabilistic pose estimation from sonar: Bayesian state estimation on the pose manifold, with uncertainty propagated coherently from sensing through to decision.

It will deliver methods for sonar-based SLAM and localization and field-validated systems that run them. Benchmarking in the wild requires ground truth that turned out not to exist, so the project also builds the algorithms and hardware rigs to capture it, along with the datasets they produce.

Stein Mixture Filtering: Advancing Bayesian Data Science Towards Real-Time Inference

2025-07-01T00:00:00+00:00

Real-time Bayesian inference remains difficult in the non-Gaussian, high-dimensional settings where data science increasingly operates. Particle filters degenerate as dimension grows, while standard variational filters impose Gaussian assumptions that misrepresent a system's true uncertainty.

This project develops variational mixture filtering to address this, fitting mixture variational families that capture non-Gaussian, multimodal posteriors while remaining cheap enough to update in real time.

It will deliver theory for the convergence of mixture flows in variational filtering, scalable filtering methods, and open-source implementations for real-time Bayesian data science.

ELBOing Stein: Variational Bayes with Stein Mixture Inference

2025-04-24T00:00:00+00:00

Stein variational gradient descent (SVGD) (Liu & Wang, 2016) performs approx- imate Bayesian inference by representing the posterior with a set of particles. However, SVGD suffers from variance collapse, i.e. poor predictions due to under- estimating uncertainty (Ba et al., 2021), even for moderately-dimensional models such as small Bayesian neural networks (BNNs). To address this issue, we gen- eralize SVGD by letting each particle parameterize a component distribution in a mixture model. Our method, Stein Mixture Inference (SMI), optimizes a lower bound to the evidence (ELBO) and introduces user-specified guides parameterized by particles. SMI extends the Nonlinear SVGD framework (Wang & Liu, 2019) to the case of variational Bayes. SMI effectively avoids variance collapse, judging by a previously described test developed for this purpose, and performs well on stan- dard data sets. In addition, SMI requires considerably fewer particles than SVGD to accurately estimate uncertainty for small BNNs. The synergistic combination of NSVGD, ELBO optimization and user-specified guides establishes a promising approach towards variational Bayesian inference in the case of tall and wide data. This is the abstract of the sample publication.

AI Implementation and Capability Development in Manufacturing: An Action Research Case

2024-01-03T00:00:00+00:00

This action research article presents a case study of a global manufacturing company deploying artificial intelligence (AI) to develop capabilities and enhance decision-making. This study explores considerations and trade-offs involved in introducing AI into daily operations, leading up to the decision to develop AI capabilities in-house or outsource them. The case study offers in-depth technical descriptions of model selection, dataset creation, model adoption, model training and evaluation while addressing organizational obstacles and decisionmaking processes. The study’s findings highlight the importance of collaboration between technical experts, business leaders, and end-users, as well as the interaction and collaboration between AI systems and human employees in the workplace. The article contributes a practical perspective on AI implementation in manufacturing, emphasizing the need to balance in-house capability development with external acquisition. Although the case study company managed to create an in-house model, factors such as implementation, debugging, data requirements, training time, and performance led to outsourcing the capabilities. However, making this informed decision required capabilities and insights that were acquired through practical work. Consequently, although inhouse development can be challenging, it can also enhance organizational capabilities and provide the necessary knowledge to make informed decisions about future development or outsourcing.

AD for an Array Language with Nested Parallelism

2022-11-13T00:00:00+00:00

We present a technique for applying reverse mode automatic differentiation (AD) on a non-recursive second-order functional array language that supports nested parallelism and is primarily aimed at efficient GPU execution. The key idea is to eliminate the need for a tape by relying on redundant execution to bring into each new scope all program variables that may be needed by the differentiated code. Efficient execution is enabled by the observation that perfectly nested scopes do not introduce re-execution and that such perfect nests can be readily produced by application of known compiler trans- formations. Our technique differentiates loops and bulk-parallel operators—e.g., map, reduce(-by-index), scan, and scatter—by specific rewrite rules and aggressively optimizes the resulting nested-parallel code. We report an evaluation that compares with established AD solutions and demonstrates competitive performance on ten common benchmarks from recent applied AD literature.

Ancestral Protein Sequence Reconstruction using a Tree-Structured Ornstein-Uhlenbeck Variational Autoencoder

2021-09-29T00:00:00+00:00

Time-efficient Bayesian Inference for a (Skewed) Von Mises Distribution on the Torus in a Deep Probabilistic Programming Language

2021-09-23T00:00:00+00:00

Probabilistic programming languages (PPLs) are at the interface between statistics and the theory of programming languages. PPLs formulate statistical models as stochastic programs that enable automatic inference algorithms and optimization. Pyro and its sibling NumPyro are PPLs built on top of the deep learning frameworks PyTorch and Jax, respectively, providing simple interfaces for inference using efficient implementations of Hamiltonian Monte Carlo (HMC), the No-U-Turn Sampler (NUTS), and Stochastic Variational Inference (SVI). The Sine von Mises distribution and its skewed variant are toroidal distributions relevant to protein bioinformatics. We demonstrate the use of the skewed Sine von Mises distribution by modeling dihedral angles of proteins using a Bayesian mixture model inferred using NUTS, exploiting NumPyro's facilities for automatic enumeration.

Linguistic Representations in Multi-task Neural Networks for Ellipsis Resolution

2018-11-01T00:00:00+00:00

Sluicing resolution is the task of identifying the antecedent to a question ellipsis. Antecedents are often sentential constituents, and previous work has therefore relied on syntactic parsing, together with complex linguistic features. A recent model instead used partial parsing as an auxiliary task in sequential neural network architectures to inject syntactic information. We explore the linguistic information being brought to bear by such networks, both by defining subsets of the data exhibiting relevant linguistic characteristics, and by examining the internal representations of the network. Both perspectives provide evidence for substantial linguistic knowledge being deployed by the neural networks.

Sluice resolution without hand-crafted features over brittle syntax trees

2018-06-01T00:00:00+00:00

Sluice resolution in English is the problem of finding antecedents of wh-fronted ellipses. Previous work has relied on handcrafted features over syntax trees that scale poorly to other languages and domains; in particular, to dialogue, which is one of the most interesting applications of sluice resolution. Syntactic information is arguably important for sluice resolution, but we show that multi-task learning with partial parsing as auxiliary tasks effectively closes the gap and buys us an additional 9% error reduction over previous work. Since we are not directly relying on features from partial parsers, our system is more robust to domain shifts, giving a 26% error reduction on embedded sluices in dialogue.