Synthesis of a Multi-Terrain Motion Controller for Robot Navigation via Adaptive Open-World Learning

This thesis introduces a real-time, open-world learning (OWL) terrain-adaptive motion control system for Wheeled Mobile Robots (WMRs), addressing autonomous navigation on unknown terrains. Central to this approach is the Extended Self-Organizing Incremental Neural Network (ESOINN+), which integrates Gaussian similarity, adaptive thresholds, ghost nodes, and batch-based clustering to classify and assimilate known and unknown terrains, achieving up to 87.9% open-world accuracy. A Multi-Terrain Controller (MTC), deployed on the Leo Rover, combines visual and inertial sensing with adaptive PWM control, demonstrating robust trajectory tracking and reduced vibrations by 41%, validating a resilient, scalable autonomy framework suitable for dynamic, real-world robotic environments.