Overview
Reinforcement learning studies agents that choose actions in an environment and learn from resulting rewards or observations. It is especially useful when decisions change what information and opportunities become available later.
What it is
An RL problem describes states, actions, transitions, and a reward or utility signal. The agent seeks a policy that balances immediate outcomes with future return, often under uncertainty and incomplete knowledge.
How it works
An agent explores by taking actions, observes consequences, and updates a value estimate or policy. Algorithms may learn from a model of the environment, direct experience, demonstrations, or feedback from humans. Simulation and offline data can reduce costly or unsafe real-world exploration.
Key concepts
- States, actions, and rewards
- Policies and value functions
- Exploration and exploitation
- Credit assignment
- Planning and model learning
- Offline and safe RL
Current research questions
- How can agents learn safely when exploration has real costs?
- How should rewards encode goals without inviting unintended behavior?
- Can policies transfer across tasks, environments, and embodiments?
- How can humans provide feedback that remains reliable at scale?
Applications
- Robotics and control
- Resource and traffic management
- Personalized tutoring
- Operations and scheduling
- Scientific experiment design
Relevant research papers
A live selection from the ScholarLens research index. Open any result to read its full paper page and use the Reading Assistant where available.
Keep exploring
Search the literature with your own question, or open a paper and read it closely with ScholarLens.