Papers
Topics
Authors
Recent
Gemini 2.5 Flash
Gemini 2.5 Flash
102 tokens/sec
GPT-4o
59 tokens/sec
Gemini 2.5 Pro Pro
43 tokens/sec
o3 Pro
6 tokens/sec
GPT-4.1 Pro
50 tokens/sec
DeepSeek R1 via Azure Pro
28 tokens/sec
2000 character limit reached

LARP: Language-Agent Role Play for Open-World Games (2312.17653v1)

Published 24 Dec 2023 in cs.AI

Abstract: Language agents have shown impressive problem-solving skills within defined settings and brief timelines. Yet, with the ever-evolving complexities of open-world simulations, there's a pressing need for agents that can flexibly adapt to complex environments and consistently maintain a long-term memory to ensure coherent actions. To bridge the gap between language agents and open-world games, we introduce Language Agent for Role-Playing (LARP), which includes a cognitive architecture that encompasses memory processing and a decision-making assistant, an environment interaction module with a feedback-driven learnable action space, and a postprocessing method that promotes the alignment of various personalities. The LARP framework refines interactions between users and agents, predefined with unique backgrounds and personalities, ultimately enhancing the gaming experience in open-world contexts. Furthermore, it highlights the diverse uses of LLMs in a range of areas such as entertainment, education, and various simulation scenarios. The project page is released at https://miao-ai-lab.github.io/LARP/.

PDF HTML Abstract

Overview

The integration of language agents into open-world games marks a significant advancement in the utilization of LLMs in interactive environments. Language Agent for Role-Playing (LARP) elevates this concept by introducing a cognitive architecture designed to simulate human-like memory and decision-making processes. This framework enables agents to adapt to diverse personalities, enhance engagement, and promote a realistic gaming experience.

Cognitive Architecture

The cognitive architecture of LARP consists of four primary components: long-term memory, working memory, memory processing, and decision-making. These elements work in unison to store experiences, process inputs from the game environment, recall relevant information, and generate coherent actions. The system incorporates cognitive psychology principles, allowing agents to emulate human memory functions such as encoding, storage, and recall. Additionally, LARP leverages a cluster of small, domain-specific LLMs that operate more efficiently than a single large-scale model without sacrificing sophistication.

Environmental Interaction

Interaction with the game environment is a core aspect of LARP's functionality. The framework facilitates the breakdown of tasks into actionable steps that agents can understand and execute. It utilizes an "Action Space" comprising APIs that represent the available actions within the game. Agents can learn new actions, which are then stored for future use, enhancing their ability to navigate and manipulate the open-world environment effectively.

Personalities

LARP agents are pre-equipped with unique personalities and backgrounds, allowing them to exhibit diverse behaviors aligned with their roles within the game world. This variety enriches the interaction potential of non-player characters, providing players with a more dynamic and immersive experience. The system uses fine-tuned LLMs representing various cultural and social perspectives to achieve this personalization, ensuring that each agent's actions and expressions are consistent with its character.

Discussion and Implications

LARP not only improves the single-user gaming experience but also offers a broader social dimension by potentially enabling multi-agent cooperation. This could lead to a more vibrant and interconnected game world, where agents can socialize and engage in complex interactions. There are challenges, such as ensuring logical consistency across multiple agents and minimizing cognitive distortion, but the development of an evaluation and feedback mechanism could help address these concerns. Conclusively, LARP could serve as a stepping stone towards more lifelike and engaging virtual environments that extend beyond entertainment into areas like education and simulations.

File Document Download Save Streamline Icon: https://streamlinehq.com PDF File Document Download Save Streamline Icon: https://streamlinehq.com Markdown Bookmark Chat Bubble Oval Streamline Icon: https://streamlinehq.com Chat (Pro)
Definition Search Book Streamline Icon: https://streamlinehq.com
References (69)
  1. Llama: Open and efficient foundation language models. arXiv preprint arXiv:2302.13971, 2023.
  2. Improving language understanding by generative pre-training. 2018.
  3. Language models are unsupervised multitask learners. OpenAI blog, 1(8):9, 2019.
  4. Language models are few-shot learners. Advances in neural information processing systems, 33:1877–1901, 2020.
  5. Training language models to follow instructions with human feedback. Advances in Neural Information Processing Systems, 35:27730–27744, 2022.
  6. OpenAI. Gpt-4 technical report, 2023.
  7. Artificial intelligence a modern approach. London, 2010.
  8. Intelligent agents: Theory and practice. The knowledge engineering review, 10(2):115–152, 1995.
  9. A survey on large language model based autonomous agents. arXiv preprint arXiv:2308.11432, 2023a.
  10. The rise and potential of large language model based agents: A survey. arXiv preprint arXiv:2309.07864, 2023.
  11. Cognitive architectures for language agents. arXiv preprint arXiv:2309.02427, 2023.
  12. Playing text-based games with common sense. arXiv preprint arXiv:2012.02757, 2020.
  13. Pre-trained language models as prior knowledge for playing text-based games. arXiv preprint arXiv:2107.08408, 2021.
  14. Reading and acting while blindfolded: The need for semantics in text game agents. arXiv preprint arXiv:2103.13552, 2021.
  15. Learning to speak and act in a fantasy text adventure game. arXiv preprint arXiv:1903.03094, 2019.
  16. Dota 2 with large scale deep reinforcement learning, 2019.
  17. Alphastar: An evolutionary computation perspective. In Proceedings of the genetic and evolutionary computation conference companion, pages 314–315, 2019.
  18. Agentsims: An open-source sandbox for large language model evaluation. arXiv preprint arXiv:2308.04026, 2023.
  19. Generative agents: Interactive simulacra of human behavior. In Proceedings of the 36th Annual ACM Symposium on User Interface Software and Technology, pages 1–22, 2023.
  20. Rolellm: Benchmarking, eliciting, and enhancing role-playing abilities of large language models. arXiv preprint arXiv:2310.00746, 2023b.
  21. Character-llm: A trainable agent for role-playing. arXiv preprint arXiv:2310.10158, 2023.
  22. Role play with large language models. Nature, pages 1–6, 2023.
  23. Chatharuhi: Reviving anime character in reality via large language model. arXiv preprint arXiv:2308.09597, 2023a.
  24. Camel: Communicative agents for” mind” exploration of large scale language model society. arXiv preprint arXiv:2303.17760, 2023b.
  25. Gamegpt: Multi-agent collaborative framework for game development. arXiv preprint arXiv:2310.08067, 2023a.
  26. Communicative agents for software development. arXiv preprint arXiv:2307.07924, 2023.
  27. Metaagents: Simulating interactions of human behaviors for llm-based task-oriented coordination via collaborative generative agents. arXiv preprint arXiv:2310.06500, 2023c.
  28. Autogen: Enabling next-gen llm applications via multi-agent conversation framework. arXiv preprint arXiv:2308.08155, 2023.
  29. Recagent: A novel simulation paradigm for recommender systems. arXiv preprint arXiv:2306.02552, 2023c.
  30. Training socially aligned language models in simulated human society. arXiv preprint arXiv:2305.16960, 2023a.
  31. Encouraging divergent thinking in large language models through multi-agent debate. arXiv preprint arXiv:2305.19118, 2023.
  32. Improving factuality and reasoning in language models through multiagent debate. arXiv preprint arXiv:2305.14325, 2023.
  33. Chateval: Towards better llm-based evaluators through multi-agent debate. arXiv preprint arXiv:2308.07201, 2023.
  34. Textworld: A learning environment for text-based games. In Computer Games: 7th Workshop, CGW 2018, Held in Conjunction with the 27th International Conference on Artificial Intelligence, IJCAI 2018, Stockholm, Sweden, July 13, 2018, Revised Selected Papers 7, pages 41–75. Springer, 2019.
  35. Interactive fiction games: A colossal adventure. In Proceedings of the AAAI Conference on Artificial Intelligence, volume 34, pages 7903–7910, 2020.
  36. Describe, explain, plan and select: Interactive planning with large language models enables open-world multi-task agents. arXiv preprint arXiv:2302.01560, 2023d.
  37. Voyager: An open-ended embodied agent with large language models. arXiv preprint arXiv:2305.16291, 2023e.
  38. Ghost in the minecraft: Generally capable agents for open-world enviroments via large language models with text-based knowledge and memory. arXiv preprint arXiv:2305.17144, 2023.
  39. Lilian Weng. Llm-powered autonomous agents. lilianweng.github.io, Jun 2023. URL https://lilianweng.github.io/posts/2023-06-23-agent/.
  40. Chatdb: Augmenting llms with databases as their symbolic memory. arXiv preprint arXiv:2306.03901, 2023.
  41. Memorybank: Enhancing large language models with long-term memory. arXiv preprint arXiv:2305.10250, 2023.
  42. Memory sandbox: Transparent and interactive memory management for conversational agents. In Adjunct Proceedings of the 36th Annual ACM Symposium on User Interface Software and Technology, pages 1–3, 2023.
  43. Tree of thoughts: Deliberate problem solving with large language models. arXiv preprint arXiv:2305.10601, 2023.
  44. Llm+ p: Empowering large language models with optimal planning proficiency. arXiv preprint arXiv:2304.11477, 2023b.
  45. React: Synergizing reasoning and acting in language models. arXiv preprint arXiv:2210.03629, 2022.
  46. Reflexion: Language agents with verbal reinforcement learning. In Thirty-seventh Conference on Neural Information Processing Systems, 2023.
  47. Chain of hindsight aligns language models with feedback. arXiv preprint arXiv:2302.02676, 3, 2023c.
  48. Plan-and-solve prompting: Improving zero-shot chain-of-thought reasoning by large language models. arXiv preprint arXiv:2305.04091, 2023f.
  49. Chain-of-thought prompting elicits reasoning in large language models. Advances in Neural Information Processing Systems, 35:24824–24837, 2022.
  50. Self-refine: Iterative refinement with self-feedback. arXiv preprint arXiv:2303.17651, 2023.
  51. Webgpt: Browser-assisted question-answering with human feedback. arXiv preprint arXiv:2112.09332, 2021.
  52. Toolformer: Language models can teach themselves to use tools. arXiv preprint arXiv:2302.04761, 2023.
  53. John E Laird. The Soar cognitive architecture. MIT press, 2019.
  54. Endel Tulving et al. Episodic and semantic memory. Organization of memory, 1(381-403):1, 1972.
  55. Henry L Roediger. Implicit memory: Retention without remembering. American psychologist, 45(9):1043, 1990.
  56. Alan Baddeley. Working memory: looking back and looking forward. Nature reviews neuroscience, 4(10):829–839, 2003.
  57. Plans and the structure of behaviour. In Systems Research for Behavioral Science, pages 369–382. Routledge, 2017.
  58. Human memory: A proposed system and its control processes. In Psychology of learning and motivation, volume 2, pages 89–195. Elsevier, 1968.
  59. George A Miller. The magical number seven, plus or minus two: Some limits on our capacity for processing information. Psychological review, 63(2):81, 1956.
  60. Arthur W Melton. Implications of short-term memory for a general theory of memory. Journal of verbal Learning and verbal Behavior, 2(1):1–21, 1963.
  61. From word models to world models: Translating from natural language to the probabilistic language of thought. arXiv preprint arXiv:2306.12672, 2023.
  62. Frederic Charles Bartlett. Remembering: A study in experimental and social psychology. Cambridge university press, 1995.
  63. Measuring and narrowing the compositionality gap in language models. arXiv preprint arXiv:2210.03350, 2022.
  64. Reconstruction of automobile destruction: An example of the interaction between language and memory. Journal of verbal learning and verbal behavior, 13(5):585–589, 1974.
  65. The wickelgren power law and the ebbinghaus savings function. Psychological Science, 18(2):133–134, 2007.
  66. Octopus: Embodied vision-language programmer from environmental feedback. arXiv preprint arXiv:2310.08588, 2023.
  67. Gorilla: Large language model connected with massive apis. arXiv preprint arXiv:2305.15334, 2023.
  68. Maybe only 0.5% data is needed: A preliminary exploration of low training data instruction tuning. arXiv preprint arXiv:2305.09246, 2023b.
  69. How abilities in large language models are affected by supervised fine-tuning data composition. arXiv preprint arXiv:2310.05492, 2023.
User Edit Pencil Streamline Icon: https://streamlinehq.com
Authors (6)
  1. Ming Yan (190 papers)
  2. Ruihao Li (4 papers)
  3. Hao Zhang (948 papers)
  4. Hao Wang (1120 papers)
  5. Zhilan Yang (2 papers)
  6. Ji Yan (4 papers)
Citations (12)
View on Semantic Scholar
Github Logo Streamline Icon: https://streamlinehq.com

GitHub

  1. LARP: LANGUAGE-AGENT ROLE PLAY FOR OPEN-WORLD GAMES
Youtube Logo Streamline Icon: https://streamlinehq.com

YouTube