【摘要】 无人移动平台是改变未来人类生产生活方式和战争形态的新质科技力量,其核心在于利用人工智能技术提升无人移动平台的自主化和智能化水平.然而,现阶段的人工智能模型参数、结构、推理链等核心要素固化,当无人移动平台面对复杂多变的对象、环境、任务以及资源受限的硬件平台,泛化和外推能力严重受限,无法满足实际应用需求.为了攻克这一技术难题,受生命智能体进化机制启发,本文提出了面向无人移动平台的自主进化学习方法,并根据进化模式的不同,将智能模型的自主进化过程划分为即时性进化、长时性进化和推理链进化3个层次;详细讨论了每个层次智能模型进化方法的技术路线和优缺点;最后,对智能模型自主进化技术在无人移动平台上的应用进行了展望与分析,并指出了现阶段自主进化学习方法存在的问题以及未来的研究方向.更多还原

【Abstract】 The unmanned mobile platform stands as a pivotal technological innovation with transformative implications for various aspects of human society, ranging from production methodologies to lifestyle choices and even the dynamics of warfare. At the heart of this transformation lies the integration of artificial intelligence(AI) technologies, aimed at augmenting the autonomy and intelligence of these platforms. However, the current landscape of AI models is marked by inherent limitations, such as fixed parameters, rigid structures, and constrained inference chains. When faced with the intricacies of complex and ever-changing objects, diverse environments, multifaceted tasks, and hardware limitations, the unmanned mobile platforms encounter severe constraints in their generalization and extrapolation capabilities. These limitations pose a significant challenge, hindering their ability to meet the diverse and dynamic demands of practical applications.In response to this challenge, this research draws inspiration from the evolutionary mechanisms observed in living organisms. A groundbreaking solution is proposed in the form of an autonomous evolutionary learning approach,meticulously tailored to address the unique requirements of unmanned mobile platforms. This innovative approach reimagines the evolution process of intelligent models across three distinct levels: immediate evolution, long-term evolution, and inference chain Autonomous evolution. By embracing different evolutionary patterns at each level, this methodology seeks to enhance the adaptability and intelligence of unmanned mobile platforms, enabling them to navigate the complexities of their operational contexts with finesse.This paper meticulously explores the technical pathways and nuances associated with evolutionary methods deployed at each level. In-depth analyses are conducted to assess the advantages and disadvantages inherent in these approaches. By dissecting the intricacies of immediate evolution, which focuses on rapid adaptability to immediate changes; long-term evolution, emphasizing sustained learning and adaptation over extended periods; and inference chain evolution,concentrating on the refinement of reasoning processes, this research illuminates the multifaceted strategies employed to bolster the intelligence of unmanned mobile platforms.Furthermore, this paper provides a comprehensive prospective analysis of the practical applications of autonomous evolutionary learning techniques within the realm of unmanned mobile platforms. It delves into the myriad ways these techniques can revolutionize sectors such as autonomous transportation, disaster response, surveillance, and military operations. Additionally, this research critically examines the existing challenges faced by autonomous evolutionary learning methods, including ethical considerations, data privacy concerns, and computational complexities.In conclusion, this extensive exploration underscores the transformative potential of autonomous evolutionary learning techniques in reshaping the landscape of unmanned mobile platforms. By addressing the limitations of current AI models and harnessing the power of evolutionary principles, these platforms can transcend their existing constraints and evolve into highly intelligent, adaptable, and context-aware entities.更多还原