Search results

(1 - 1 of 1)

Title

EVALUATING INTEGRITY FOR MOBILE ROBOT LOCALIZATION SAFETY

Creator

Duenas Arana, Guillermo

Date

2019

Description

Precise localization is paramount for autonomous navigation. Localization errors are not only dangerous by themselves, but can also mislead...
Show morePrecise localization is paramount for autonomous navigation. Localization errors are not only dangerous by themselves, but can also mislead other dependent systems into moving to a hazardous location. Unfortunately, the problem of quantifying robot localization safety is only sparsely addressed in the robotics literature, and most robotics algorithms still quantify pose estimation performance using a covariance matrix or particle spread, which only accounts for nominal sensor errors. This is insufficient for life- and mission-critical applications, such as autonomous vehicles and other co-robots, where ignoring sensor or sensor or processing faults can lead to catastrophic localization errors. Thus, other methods must be employed to ensure safety.In response, this research leverages prior work in aviation integrity monitoring to tackle the more challenging case of evaluating localization safety in mobile robots. In contrast to aviation applications, that heavily rely on the Global Navigation Satellite System (GNSS) for localization, robots often operate in complex, GNSS-denied environments that require a more sophisticated sensor suite to ensure localization safety. Localization integrity risk is the probability that a robot's pose estimate lies outside pre-defined acceptable limits while no alarm is triggered. In this work, the integrity risk is rigorously upper bounded by accounting for both nominal noise and other non-nominal sensor faults, resulting in a safe upper bound on the localization integrity risk.The main contribution of this dissertation is the design and evaluation of a sequential integrity monitoring methodology applicable to mobile robot localization algorithms that use feature extraction and data association. First, faults introduced during the feature extraction and data association processes are distinguished, and the probability of the latter is rigorously upper bounded using analytical methods. The impact of faults in the estimate error's and fault detector's distributions is then determined to quantify integrity risk, which is evaluated under the worst-possible fault combination. To determine the impact of previous faults without a boundlessly growing number of fault hypotheses, this dissertation presents a novel method that uses a preceding time window to build a limited set of hypotheses and a prior estimate bias to account for faults occurring before the start of the time window. The proposed methodology is applicable to Kalman Filter and fixed-lag smoothing localization. Simulated and experimental results are presented to validate the methodology.
Show less