Task and Motion Planning in Neuro-Symbolic Robotics: From Individual Agents to Multi-Agent Systems

Autonomous robotic systems operating in continuous and highly dynamic environments require reliable and consistent decision-making over long horizons. To address these challenges, this thesis develops hierarchical decision-making frameworks that decompose complex planning problems into multiple layers of abstraction, spanning from high-level task reasoning to low-level motion control. Such hierarchical structures enhance scalability, interpretability, and modularity, enabling robust reasoning across diverse temporal and spatial resolutions. The thesis focuses on two representative case studies: robotic object manipulation in cooperative and collaborative scenarios, and autonomous robot racing in competitive settings. Key contributions include: (i) methods for the automatic acquisition of taskoriented abstractions from unstructured data, enabling interpretable high-level planning, (ii) the development of temporal planning frameworks that coordinate actions over long horizons, and (iii) the formulation of strategic decision-making mechanisms for competitive multi-agent scenarios using non-cooperative game-theoretic reasoning. By leveraging neuro-symbolic reasoning and foundation-model-driven abstractions for cooperative and collaborative scenarios, and non-cooperative game-theoretic planning for competitive scenarios, this research demonstrates that robots can achieve robust, interpretable, and adaptive behavior across diverse multi-agent environments. The proposed frameworks advance the state of the art in autonomous decision-making, providing a foundation for future research in scalable, generalizable, and strategically capable robotic systems.