This dissertation contributes to the collaborative multi-robot systems literature, which is predominantly hindered by reliance on expensive sensors and predefined learning models, scalability issues, high communication costs, and difficulties adapting to environmental changes. The research proposes innovative algorithms and strategies for improving localization and exploration capabilities for multi-robot collaboration in ad hoc networks.

We introduce an online localization algorithm that enhances accuracy and efficiency by combining wireless sensor nodes (WSN) and mobile robots, leveraging Radio Signal Strength Indicator (RSSI). A relative localization technique is proposed to enhance this algorithm, effectively working in scenarios lacking infrastructure information. It combines graph optimization and Gaussian process regression, surpassing traditional model-based methods. A mechanism is designed to provide resilience to robotic systems, making localization solutions failure-tolerant, thus ensuring high positional accuracy in real-world applications.

Additionally, we develop a learning-based framework for simultaneous localization and adaptive exploration by generating virtual maps for efficient exploration in dynamic environments. This methodology is founded on Bayesian inference over the Gaussian probability distribution of wireless signals. A strategy is presented for coordinated map exploration that integrates Q-learning for efficient path planning. We devise a feature-matching map-merger strategy to create a consistent map from sparse maps collected from collaborating robots. An enhanced reinforcement learning technique is deployed for maze exploration, providing superior efficiency in challenging environments by increasing coverage and reducing overlap. We also investigate an integrated framework for simultaneous exploration and localization, which is crucial for robotic operations in dynamic environments.

The research comprehensively addresses the critical issues facing current CMRS by introducing novel algorithms for improved localization and exploration. These strategies demonstrate significant improvements in scalability, communication, resilience to node failure, and adaptability to dynamic environments. The solutions have proven effective in both theoretical models and real-world applications, strengthening their viability. By setting new standards in autonomous robotic collaboration, this dissertation provides a strong foundation for future research in the field.