Improving Long Term Accuracy of Visual Localization in Urban Environment

Abstract

Localization remains a pivotal aspect of mobile robotics, with robots required to discern their position by comparing sensor inputs against a pre-established environmental map. Notably, environmental shifts over time can diminish the reliability of these localization efforts. Addressing this challenge, our study introduces two interventions: dynamic object masking and CNN model fine-tuning, both scrutinized through extensive real-world experiments involving a robot operating continuously over a four-month span. Such long-duration testing of mobile robot localization in fluctuating environments is scarcely reported in existing literature where most focus on localization system usage on one-shot or on a short period of time. Additionally, robustness of localization can only be observed in real world usage over a long term. Our findings reveal that while fine-tuning the CNN model may not drastically enhance immediate testing environment accuracy, it significantly bolsters the system's robustness and accuracy, especially when adapted to robots equipped with varying sensor technologies. This work not only contributes to the field by providing rare empirical evidence of extended operational challenges and solutions but also unveils deeper understandings pivotal for advancing mobile robot localization under the inevitable condition of environmental change.