This paper presents a comprehensive survey on Role-Playing Language Agents (RPLAs), exploring their evolution, current progress, and future prospects. RPLAs are specialized AI systems designed to simulate assigned personas, leveraging the advanced capabilities of large language models (LLMs), including in-context learning, instruction following, and social intelligence. These systems can mimic a wide range of personas, from historical figures and fictional characters to real-life individuals, and have catalyzed numerous AI applications, such as emotional companions, interactive video games, personalized assistants, and digital clones. The paper categorizes personas into three types: 1) Demographic Persona, which leverages statistical stereotypes; 2) Character Persona, focused on well-established figures; and 3) Individualized Persona, customized through ongoing user interactions for personalized services. It provides a detailed overview of current methodologies for RPLAs, covering data sourcing, agent construction, and evaluation for each persona type. The paper also discusses the fundamental risks, existing limitations, and future prospects of RPLAs, as well as their practical applications in AI. Recent developments in LLMs have significantly facilitated the emergence of RPLAs. LLMs are capable of producing a compelling sense of human likeness and can be regarded as superpositions of beliefs and personas. With alignment training, LLMs can adhere to the instruction of persona role-playing, including replicating their knowledge, linguistic and behavior patterns, and even underlying personalities. These capabilities enable RPLAs to both mimic the personas as prompted in the contexts or harness their inherent parametric knowledge for widely-recognized demographics or characters. The paper also explores the application of RPLAs in various domains, including single-agent and multi-agent systems, where assigning specific demographics can enhance task-solving performance. In multi-agent systems, diverse personas can be assigned to agents to cultivate distinct societal dynamics, leading to improved strategies for cooperative problem-solving and breakthroughs in complex domains. Additionally, RPLAs have demonstrated remarkable capabilities in simulating nuanced, human-like interactions across various environments, including gaming and social simulations. The paper concludes by discussing the evaluation of RPLAs, considering their character-independent capabilities and character fidelity. It highlights the importance of assessing how well RPLAs can simulate the intended characters, including their linguistic style, knowledge, personality, and thought processes. The evaluation methods include automatic evaluation with ground truth, automatic evaluation without ground truth, multi-choice questions, and human evaluation. The paper emphasizes the need for further research to address the challenges and limitations of RPLAs, ensuring their safe and effective use in various applications.This paper presents a comprehensive survey on Role-Playing Language Agents (RPLAs), exploring their evolution, current progress, and future prospects. RPLAs are specialized AI systems designed to simulate assigned personas, leveraging the advanced capabilities of large language models (LLMs), including in-context learning, instruction following, and social intelligence. These systems can mimic a wide range of personas, from historical figures and fictional characters to real-life individuals, and have catalyzed numerous AI applications, such as emotional companions, interactive video games, personalized assistants, and digital clones. The paper categorizes personas into three types: 1) Demographic Persona, which leverages statistical stereotypes; 2) Character Persona, focused on well-established figures; and 3) Individualized Persona, customized through ongoing user interactions for personalized services. It provides a detailed overview of current methodologies for RPLAs, covering data sourcing, agent construction, and evaluation for each persona type. The paper also discusses the fundamental risks, existing limitations, and future prospects of RPLAs, as well as their practical applications in AI. Recent developments in LLMs have significantly facilitated the emergence of RPLAs. LLMs are capable of producing a compelling sense of human likeness and can be regarded as superpositions of beliefs and personas. With alignment training, LLMs can adhere to the instruction of persona role-playing, including replicating their knowledge, linguistic and behavior patterns, and even underlying personalities. These capabilities enable RPLAs to both mimic the personas as prompted in the contexts or harness their inherent parametric knowledge for widely-recognized demographics or characters. The paper also explores the application of RPLAs in various domains, including single-agent and multi-agent systems, where assigning specific demographics can enhance task-solving performance. In multi-agent systems, diverse personas can be assigned to agents to cultivate distinct societal dynamics, leading to improved strategies for cooperative problem-solving and breakthroughs in complex domains. Additionally, RPLAs have demonstrated remarkable capabilities in simulating nuanced, human-like interactions across various environments, including gaming and social simulations. The paper concludes by discussing the evaluation of RPLAs, considering their character-independent capabilities and character fidelity. It highlights the importance of assessing how well RPLAs can simulate the intended characters, including their linguistic style, knowledge, personality, and thought processes. The evaluation methods include automatic evaluation with ground truth, automatic evaluation without ground truth, multi-choice questions, and human evaluation. The paper emphasizes the need for further research to address the challenges and limitations of RPLAs, ensuring their safe and effective use in various applications.