NANO-SLAM : Natural Gradient Gaussian Approximation for Vehicle SLAM

2025-04-27

AI Summary

To address degraded SLAM localization accuracy of autonomous vehicles in GPS-denied environments (e.g., urban canyons, tunnels), this paper proposes a novel linearization-free Rao-Blackwellized particle filter (RBPF) framework. Departing from conventional EKF/UKF-based posterior linearization approximations, it introduces natural gradient descent—applied for the first time—to optimize parameters of Gaussian variational distributions, enabling geometry-aware, unbiased approximation of the nonlinear posterior. Furthermore, it enhances particle efficacy and convergence speed by optimizing the proposal distribution. Evaluated on the Sydney Victoria Park real-world dataset, the method reduces positioning error by over 50% while maintaining nearly identical computational overhead. Key contributions include: (i) the first integration of natural gradient optimization into the RBPF sampling mechanism; (ii) robust nonlinear state estimation without Jacobian or Hessian computation; and (iii) significantly improved long-term localization accuracy and stability under GPS-denied conditions.

Technology Category

• Intelligent Robots: State Estimation
• Search and Optimization: Sampling/Simulation-based Search
• Reasoning under Uncertainty: Stochastic Optimization

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Graph Algorithms and Modeling for the Web: Efficient manipulation of static and dynamic Web-related graphs
• User Modeling, Personalization and Recommendation: On-Device user modeling, personalization, and recommendation

Abstract

Accurate localization is a challenging task for autonomous vehicles, particularly in GPS-denied environments such as urban canyons and tunnels. In these scenarios, simultaneous localization and mapping (SLAM) offers a more robust alternative to GPS-based positioning, enabling vehicles to determine their position using onboard sensors and surrounding environment's landmarks. Among various vehicle SLAM approaches, Rao-Blackwellized particle filter (RBPF) stands out as one of the most widely adopted methods due to its efficient solution with logarithmic complexity relative to the map size. RBPF approximates the posterior distribution of the vehicle pose using a set of Monte Carlo particles through two main steps: sampling and importance weighting. The key to effective sampling lies in solving a distribution that closely approximates the posterior, known as the sampling distribution, to accelerate convergence. Existing methods typically derive this distribution via linearization, which introduces significant approximation errors due to the inherent nonlinearity of the system. To address this limitation, we propose a novel vehicle SLAM method called extit{N}atural Gr extit{a}dient Gaussia extit{n} Appr extit{o}ximation (NANO)-SLAM, which avoids linearization errors by modeling the sampling distribution as the solution to an optimization problem over Gaussian parameters and solving it using natural gradient descent. This approach improves the accuracy of the sampling distribution and consequently enhances localization performance. Experimental results on the long-distance Sydney Victoria Park vehicle SLAM dataset show that NANO-SLAM achieves over 50% improvement in localization accuracy compared to the most widely used vehicle SLAM algorithms, with minimal additional computational cost.

Problem

• Improving vehicle localization in GPS-denied environments
• Reducing approximation errors in SLAM sampling distribution
• Enhancing accuracy of Rao-Blackwellized particle filter SLAM

Innovation

• Uses natural gradient descent for optimization
• Models sampling distribution as Gaussian parameters
• Avoids linearization errors in SLAM