From Theory to Practice: Achieving Reliable Robot Autonomy through Symbol Grounding

In an era where robotic autonomy is becoming increasingly pivotal across various sectors, the symbol grounding problem remains a significant barrier to achieving reliable, context-aware automation. This paper presents novel frameworks to enhance robot autonomy by integrating symbol grounding into autonomous systems, specifically focusing on robot manipulation tasks. We first introduce a method for programming robots using behavior trees, derived from single demonstrations, which embeds symbolic knowledge into robot operations, enabling adaptability. Second, we discuss a framework that integrates spatial scene graphs with BTs to improve task execution through an enhanced understanding of spatial relationships and object interactions, which is crucial for dynamic and cluttered environments. Lastly, we present a neuro-symbolic approach for failure detection and diagnosis in robotic systems. This approach leverages the synergy between symbolic reasoning and neural network capabilities to detect and diagnose operational failures accurately, addressing the critical need for reliability in automated systems. The preliminary evaluation results demonstrate advancements in the programming, execution, and failure detection of robot manipulation tasks, paving the way for more adaptive and intelligent robotic systems in complex real-world applications.

Keywords: Autonomous robots, Reliable robot autonomy, Robot programming by demonstration, Scene graphs, Behavior trees, Failure detection and diagnosis.