Particle Filter Enhanced by CNN-LSTM for Localization of Autonomous Robot in Indoor Environment

Abstract

Particle filters play a crucial role in indoor localization tasks where GPS or similar sensors are unavailable, owing to their adaptability to dynamic environments and contribution to precise navigation. However, despite their effectiveness in non-Gaussian and non-linear settings, particle filters face challenges such as issues with particle initialization and recovery from scenarios like robot kidnapping. These limitations impede the complete autonomy of robots, a critical aspect of their functionality. This research presents a novel solution that integrates particle filters with a CNN-LSTM network to address these challenges. Leveraging the time-series image processing capabilities of CNN-LSTMs, this architecture aids in particle filter initialization and facilitates recovery from challenging situations. The integration enhances the overall performance and autonomy of robots, making them more efficient in indoor navigation tasks.