Modeling Perceptual Aliasing in SLAM via Discrete-Continuous Graphical Models

Pierre-Yves Lajoie, Siyi Hu, Giovanni Beltrame, Luca Carlone

Abstract

Perceptual aliasing is one of the main causes of failure for Simultaneous Localization and Mapping (SLAM) systems operating in the wild. Perceptual aliasing is the phenomenon where different places generate a similar visual (or, in general, perceptual) footprint. This causes spurious measurements to be fed to the SLAM estimator, which typically results in incorrect localization and mapping results. The problem is exacerbated by the fact that those outliers are highly correlated, in the sense that perceptual aliasing creates a large number of mutually-consistent outliers. Another issue stems from the fact that most state-of-the-art techniques rely on a given trajectory guess (e.g., from odometry) to discern between inliers and outliers and this makes the resulting pipeline brittle, since the accumulation of error may result in incorrect choices and recovery from failures is far from trivial. This work provides a unified framework to model perceptual aliasing in SLAM and provides practical algorithms that can cope with outliers without relying on any initial guess. We present two main contributions. The first is a Discrete-Continuous Graphical Model (DC-GM) for SLAM: the continuous portion of the DC-GM captures the standard SLAM problem, while the discrete portion describes the selection of the outliers and models their correlation. The second contribution is a semidefinite relaxation to perform inference in the DC-GM that returns estimates with provable sub-optimality guarantees. Experimental results on standard benchmarking datasets show that the proposed technique compares favorably with state-of-the-art methods while not relying on an initial guess for optimization.

Introduction

Simultaneous Localization and Mapping (SLAM) is a fundamental capability for autonomous robots, enabling them to build maps of unknown environments while tracking their own location. A critical component of SLAM is "loop closure," where a robot recognizes a previously visited location to correct trajectory drift. However, real-world environments often exhibit perceptual aliasing—the phenomenon where distinct locations (e.g., identical doors in a hallway or repeating shelf patterns in a warehouse) produce similar sensor signatures.

Perceptual aliasing introduces two major challenges for standard SLAM solvers. First, it generates correlated outliers, where multiple false measurements are mutually consistent, misleading the robot into believing it has found a valid but incorrect loop closure. Second, most robust SLAM algorithms suffer from the "initial guess" trap, requiring a reasonably accurate starting estimate of the robot's path to distinguish inliers from outliers. If initial odometry is poor, these systems often fail catastrophically.

The Discrete-Continuous Graphical Model (DC-GM)

To address these challenges, this work introduces the Discrete-Continuous Graphical Model (DC-GM). This framework explicitly models the physical process of perceptual aliasing rather than merely attempting to mitigate its effects through heavy-tailed error distributions.

The DC-GM framework combines discrete variables for outlier selection with continuous variables for pose estimation.

As shown in Figure 1, the model consists of two layers:

  • Continuous Variables: These represent the standard SLAM problem, capturing the 2D or 3D poses (positions and orientations) of the robot.
  • Discrete Variables: These are binary "switches" associated with each loop closure. They determine whether a measurement is accepted as an inlier or rejected as an outlier.
  • Correlation Modeling: The discrete layer utilizes a Markov Random Field (MRF) structure to model the positive correlation between outliers. This allows the system to recognize that a cluster of measurements, while internally consistent, may collectively be outliers if they conflict with the broader map.

Inference via Semidefinite Relaxation

Solving the resulting non-convex optimization problem, which involves a mix of discrete and continuous variables, is traditionally difficult. The authors derive a Semidefinite Relaxation (SDP) that transforms the problem into a convex one. This approach provides global optimality guarantees and, crucially, makes the system initialization-free, allowing it to succeed even when no reliable initial trajectory guess is available.

Experimental Evaluation and Results

The DC-GM approach was evaluated using synthetic datasets (e.g., Manhattan World) and real-world benchmarks including MIT's CSAIL and Freiburg's FR079/FRH datasets. To simulate extreme perceptual aliasing, these datasets were "spoiled" with clusters of false loop closures.

Performance comparison showing the percentage of outliers rejected by DC-GM variants as the total percentage of outliers increases.

Key results include:

  • Superior Robustness: In simulations with up to 50% correlated outliers, the DC-GM approach successfully rejected all false measurements (Figure 5), outperforming state-of-the-art methods like Vertigo or Dynamic Covariance Scaling (DCS).
  • Accuracy in Real-World Scenarios: On the CSAIL dataset, DC-GM maintained a low average translation error of ~0.04m, whereas competitors often exceeded 1.4m due to an inability to handle correlated outliers.
  • Initialization Independence: The system successfully optimized maps starting from zero information, a scenario where traditional local search algorithms (like Gauss-Newton) typically fail.

Conclusion and Future Work

The DC-GM framework provides a mathematically rigorous way to handle the complexities of perceptual aliasing in SLAM. By modeling the correlation of outliers and utilizing SDP relaxation, it achieves unprecedented robustness without requiring a good initial guess. Future work aims to improve computational efficiency, as solving SDPs is currently too slow for real-time applications. Transitioning from batch processing to incremental updates and extending rigorous testing to 3D environments are also key areas for development.

Summary updated

Abstract

Perceptual aliasing is one of the main causes of failure for Simultaneous Localization and Mapping (SLAM) systems operating in the wild. Perceptual aliasing is the phenomenon where different places generate a similar visual (or, in general, perceptual) footprint. This causes spurious measurements to be fed to the SLAM estimator, which typically results in incorrect localization and mapping results. The problem is exacerbated by the fact that those outliers are highly correlated, in the sense that perceptual aliasing creates a large number of mutually-consistent outliers. Another issue stems from the fact that most state-of-the-art techniques rely on a given trajectory guess (e.g., from odometry) to discern between inliers and outliers and this makes the resulting pipeline brittle, since the accumulation of error may result in incorrect choices and recovery from failures is far from trivial. This work provides a unified framework to model perceptual aliasing in SLAM and provides practical algorithms that can cope with outliers without relying on any initial guess. We present two main contributions. The first is a Discrete-Continuous Graphical Model (DC-GM) for SLAM: the continuous portion of the DC-GM captures the standard SLAM problem, while the discrete portion describes the selection of the outliers and models their correlation. The second contribution is a semidefinite relaxation to perform inference in the DC-GM that returns estimates with provable sub-optimality guarantees. Experimental results on standard benchmarking datasets show that the proposed technique compares favorably with state-of-the-art methods while not relying on an initial guess for optimization.
BibTeX
@article{lajoie2019modeling,
  author = {Pierre-Yves Lajoie and Siyi Hu and Giovanni Beltrame and Luca Carlone},
  title = {{Modeling Perceptual Aliasing in SLAM via Discrete-Continuous Graphical Models}},
  journal = {IEEE Robotics Autom. Lett.},
  year = {2019},
  url = {https://openreview.net/forum?id=Htu5yQPgqg},
}
Figure