This paper presents a comprehensive survey of large language model (LLM)-based autonomous agents, discussing their construction, applications, and evaluation. Autonomous agents are systems that perceive and act in their environment to achieve their goals. Traditional agents rely on simple policies, but LLMs, with their vast knowledge, enable more human-like decision-making. The paper proposes a unified framework for LLM-based agents, covering memory, planning, and action modules. The memory module stores and retrieves information, while the planning module enables agents to reason and plan. The action module translates decisions into specific actions. The paper also discusses various strategies for agent construction, including profiling, memory, and planning. Applications of LLM-based agents span social science, natural science, and engineering. Evaluation strategies include subjective and objective methods. Challenges and future directions are identified, emphasizing the need for more comprehensive memory and planning capabilities. The survey highlights the potential of LLM-based agents in achieving human-like intelligence and their applications in diverse domains.This paper presents a comprehensive survey of large language model (LLM)-based autonomous agents, discussing their construction, applications, and evaluation. Autonomous agents are systems that perceive and act in their environment to achieve their goals. Traditional agents rely on simple policies, but LLMs, with their vast knowledge, enable more human-like decision-making. The paper proposes a unified framework for LLM-based agents, covering memory, planning, and action modules. The memory module stores and retrieves information, while the planning module enables agents to reason and plan. The action module translates decisions into specific actions. The paper also discusses various strategies for agent construction, including profiling, memory, and planning. Applications of LLM-based agents span social science, natural science, and engineering. Evaluation strategies include subjective and objective methods. Challenges and future directions are identified, emphasizing the need for more comprehensive memory and planning capabilities. The survey highlights the potential of LLM-based agents in achieving human-like intelligence and their applications in diverse domains.