Utilizing Foundation Models to Enhance Autonomous Robotic Systems in Dynamic and Unstructured Environments

Autonomous robotics in unstructured environments,such as disaster zones and construction sites, presents significantchallenges due to unpredictability and complexity. Traditionalrobotic systems often struggle with adaptability and generalization in these settings. Recent advancements in foundationmodels, including Large Language Models (LLMs) and LargeVision Models (LVMs), offer promising solutions by enhancingperception, decision-making, and interaction capabilities. Thispaper explores the integration of foundation models into autonomous robotics, systematically reviewing current applications,identifying challenges, and proposing future research directions.We analyze the impact of these models on various robotic tasks,assess the current level of autonomy achieved, and envisionscenarios for fully autonomous operations. Our findings indicatethat while foundation models significantly improve cognitivetasks, their application in physical interactions remains nascent.This study serves as a comprehensive benchmark for futureadvancements in autonomous robotics within dynamic and unstructured environments