Jialong Sun

Jialong Sun

My first blog: From CV to Agentic RL

1 minute read

Published:

This is my first blog. I’m trying to use this way to study latest reseach and industrial work as much as I can. My current research interests include agentic RL. Since most of my time has been devoted to CV in the last two years, it is a challenge. But I will take it.

First, I want to explore ‘what’. What is agent? What is RL? what is Agentic RL? What is their relationship?

Agent system
Figure 1. What is agent? (from [1])
RL
Figure 2. What is RL? (from ChatGPT)
Agentic RL Framework
Figure 3. What is Agentic RL? (from [2])

To put it simple, an agent is an entity which can use tools and take actions to achieve its goals in an environment. A more formal definition can be seen in [3]. RL is the methodology to train agents. Agentic RL focuses on how RL empowers LLM-based agents in dynamic environments[2]. As shown in Figure 4, I draw a diagram to describe their relationships in an intuitive way.

Relationship
Figure 4. What is their relationship?

In the next blog, I will dive into their components.

References

  1. Weng, Lilian. (Jun 2023). “LLM-powered Autonomous Agents”. Lil’Log. https://lilianweng.github.io/posts/2023-06-23-agent/.
  2. Zhang, Guibin, et al. "The Landscape of Agentic Reinforcement Learning for LLMs: A Survey." arXiv preprint arXiv:2509.02547 (2025).
  3. Wang, Hongru, et al. "Toward a Theory of Agents as Tool-Use Decision-Makers." arXiv preprint arXiv:2506.00886 (2025).