Autonomous Grasping Control via Deep LLM in Aerospace Robotics

Abstract

The rapid advancement of aerospace robotics, driven by the integration of large language models (LLMs) and deep reinforcement learning (DRL), has enabled transformative solutions for autonomous grasping and manipulation in extraterrestrial environments. This research presents a novel framework combining LLM-powered contextual reasoning with robust robotic control, addressing the unique challenges of microgravity, environmental uncertainty, and dynamic object manipulation. By leveraging YOLO-based object detection, TRPO for efficient inverse dynamics, and reinforcement learning within continuous action spaces, the proposed system achieves precise decision-making and high adaptability. Simulations validate the methodology, showcasing significant improvements in grasping accuracy and trajectory tracking compared to state-of-the-art techniques. The results underscore the potential of hybrid AI-driven approaches in advancing robotic capabilities for space exploration and operations, setting a new benchmark for autonomy, precision, and resilience in aerospace robotics.