Aim

The central aim of this research is to investigate how foundation models can be effectively leveraged to enhance decision-making capabilities in single autonomous agents operating within complex, real-world environments.

Objectives

1.    Explore and develop effective techniques for grounding foundation models within the context of sequential decision-making.  
2.    Design and evaluate innovative decision-making architectures that leverage the strengths of foundation models.  
3.    Investigate techniques for making the decision-making process of agents based on foundation models more transparent and interpretable.  
4.    Demonstrate the effectiveness of the proposed approach in enhancing decision-making performance compared to traditional methods.

Description

Foundational models have demonstrated remarkable proficiency in tasks such as language understanding [1], image generation [2], and even protein folding [3]. While these foundational models hold great potential, it is not clear how they can be used to enhance the decision making capability of autonomous agents, in complex and dynamic environments.  
The proposed research seeks to bridge the gap between the capabilities of foundation models and the sequential decision-making of agents in single-agent and multi-agent scenarios [4]. Particularly, we are interested in how reinforcement learning [5] and planning can be utilised in combination with foundational models to enable efficient adaptation to novel situations. By tackling these challenges, this research aims to unlock the full potential of foundation models for creating autonomous agents that can navigate complex tasks, make intelligent choices, and exhibit robust performance in the face of uncertainty.

References

[1] OpenAI. GPT-4 Technical Report. Technical report, 2023.
[2] Alec Radford, Jong Wook Kim, Chris Hallacy, Aditya Ramesh, Gabriel Goh, Sandhini, Agarwal, Girish Sastry, Amanda Askell, Pamela Mishkin, Jack Clark, Gretchen Krueger, and Ilya Sutskever. Learning Transferable Visual Models From Natural Language Supervision. In International conference on machine learning, pages 8748–8763. PMLR, 2021.
[3] Yang, Z., Zeng, X., Zhao, Y. et al. AlphaFold2 and its applications in the fields of biology and medicine. Sig Transduct Target Ther 8, 115 (2023). https://doi.org/10.1038/s41392-023-01381-z
[4] Albrecht, S. V., Christianos, F., & Schäfer, L. (2024). Multi-agent reinforcement learning: Foundations and modern approaches. MIT Press.
[5] Sutton, R. S., Barto, A. G. (2018 ). Reinforcement Learning: An Introduction. The MIT Press.