Towards Building Autonomous AI Agents and Robots for Open World Environments.

The shift of AI agents from controlled laboratory environments to real-world applications, such as autonomous vehicles and service robots, demands robust algorithms for navigating the intricacies of open-world scenarios. While traditional AI agents show proficiency in predictable, closed-world settings, their performance often diminishes in the dynamic and unforeseen conditions of real-world environments. My dissertation focuses on developing methods, frameworks, and domains that push the boundaries of open-world problem-solving in AI agents and robots. The central thesis question explores how AI agents can rapidly learn and adapt in open-world settings while tackling long-horizon, complex tasks. My work proposes integrative frameworks that combine reinforcement learning with symbolic planning, enabling on-the-fly adaptation of agents. Furthermore, we also propose environments designed for developing and assessing agent architectures adept at handling novelty. These advancements in open-world learning are pivotal in enhancing adaptability, speed, and robustness in AI agents and robots, laying a