A Probabilistic-based Drift Correction Module for Visual Inertial SLAMs

TL;DR

Aiming to reduce dead-reckoning drift in SLAM/VIO, the paper introduces a probabilistic drift-correction module that treats SLAM-position estimates as stochastic, seeking the mode of a joint multivariate distribution built from motion priors and geospatial information. The methodology defines Gaussian random variables $X_1$,$X_2$,$X_3$,$X_4$ with means $(\mu_1=\alpha_t,\mu_2=m_t,\mu_3=0,\mu_4=0)$ and a joint density $P=\prod_{i=1}^4 P(X_i|\mu_i,\sigma_i)$, and solves for the mode by minimizing the weighted negative log-likelihood $-\sum_{i=1}^4 w_i \ln P(X_i|\mu_i,\sigma_i)$ via gradient descent, incorporating GIS priors via a distance-transform implicit function. It is designed as a plug-in module compatible with any SLAM/VIO backend and evaluated with VINS-MONO to demonstrate drift reduction. Experimental results show drift decreasing by about $10\times$ to $20\times$ on long traverses, with some instability requiring ablation of weights $w_i$ and uncertainties $\sigma_i$. The work offers a practical, model-based approach to tightening long-range localization in GPS-denied environments and facilitates integration into existing SLAM workflows.

Abstract

Positioning is a prominent field of study, notably focusing on Visual Inertial Odometry (VIO) and Simultaneous Localization and Mapping (SLAM) methods. Despite their advancements, these methods often encounter dead-reckoning errors that leads to considerable drift in estimated platform motion especially during long traverses. In such cases, the drift error is not negligible and should be rectified. Our proposed approach minimizes the drift error by correcting the estimated motion generated by any SLAM method at each epoch. Our methodology treats positioning measurements rendered by the SLAM solution as random variables formulated jointly in a multivariate distribution. In this setting, The correction of the drift becomes equivalent to finding the mode of this multivariate distribution which jointly maximizes the likelihood of a set of relevant geo-spatial priors about the platform motion and environment. Our method is integrable into any SLAM/VIO method as an correction module. Our experimental results shows the effectiveness of our approach in minimizing the drift error by 10x in long treverses.

Figures (3)

• Figure 1: Implicit function representation of vector data through the inverse of the Distance Transform, depicting traversable paths with probabilities favoring center paths and lower values off-center. Maximum allowed distances are truncated akin to road width constraints in GIS-defined traversable paths.
• Figure 2: Spatial probabilities at $\Gamma(t)$ where red and blue denote high and low probability values, respectively. Random variables (a) motion magnitude $X_2$, (b) motion angle $X_1$ (c) heading preservation $X_4$.
• Figure 3: Comparison of reference, proposed, and SLAM trajectories (blue, red, and green) with start and end points highlighted (orange and black), alongside spatial distributions generated by the proposed method on the right had side.