Robust Monte-Carlo Localization Using Adaptive Likelihood Models

Abstract

In probabilistic mobile robot localization, the development of the sensor model plays a crucial role as it directly influences the efficiency and the robustness of the localization process. Sensor models developed for particle filters compute the likelihood of a sensor measurement by assuming that one of the particles accurately represents the true location of the robot. In practice, however, this assumption is often strongly violated, especially when using small sample sets or during global localization. In this paper we introduce a novel, adaptive sensor model that explicitly takes the limited representational power of particle filters into account. As a result, our approach uses smooth likelihood functions during global localization and more peaked functions during position tracking. Experiments show that our technique significantly outperforms existing, static sensor models.

Adaptive Full Scan Model for Range Finders in Dynamic Environments

“…Range finder sensor models can also be classified according to whether they use discrete geometric grids [4][5][6][7][8] or continuous geometric models [9][10][11]. Moravec [8] proposed non-Gaussian measurement densities over a discrete grid of possible distances measured by sonar; the likelihood of the measurements has to be computed for all possible positions of the mobile robot at a given time.…”

“…An analogous mixture adds two more physical causes: a sensor failure and an unknown cause resulting in a 'max-range' measurement and a 'random' measurement, respectively [9,10]. While [9,11] use a continuous model, [10] presents the discrete analog of the mixture, taking into account the limited resolution of the range sensor. [11] extend the basic mixture model for use in Monte Carlo localization.…”

“…While [9,11] use a continuous model, [10] presents the discrete analog of the mixture, taking into account the limited resolution of the range sensor. [11] extend the basic mixture model for use in Monte Carlo localization. To overcome problems due to the combination of the limited representational power and the peaked likelihood of the accurate range finder, they propose an adaptive likelihood model.…”

Gaussian mixture models for probabilistic localization

“…In practice, this problem is dealt with by sub-sampling of measurements [18], by introducing minimal likelihoods for beams, or by other means of regularization of the resulting likelihoods, see e.g.Arulampalam et al [1]. In our previous work [12], we addressed this problem by adapting the "peakedness" of the beam model using learned heuristics. In other previous work [13], [15], we introduced scan-based likelihood models.…”

“…In contrast to former approaches [8], [12], [13], [15] which modeled the likelihood functions as unimodal distributions for single beams or entire scans we now consider to model each beam independently as a mixture of K Gaussian distributions [16]. In such a mixture model, the likelihood of the i-th beam of z t becomes…”