concept

On-Policy Learning

On-policy learning is a reinforcement learning approach where an agent learns the value of the policy it is currently following, using data generated by that same policy. It involves updating the policy based on actions taken during exploration, ensuring that the learning process aligns with the behavior policy. This method is commonly used in algorithms like SARSA (State-Action-Reward-State-Action) and is contrasted with off-policy learning, which can learn from data generated by different policies.

Also known as: On-policy, On-policy RL, On-policy reinforcement learning, On-policy methods, On-policy algorithms
🧊Why learn On-Policy Learning?

Developers should learn on-policy learning when building reinforcement learning systems that require stable and consistent policy updates, such as in robotics control, game AI, or real-time decision-making applications. It is particularly useful in scenarios where exploration must be safe and predictable, as it avoids the risks associated with learning from potentially suboptimal or divergent policies, making it suitable for environments with high stakes or continuous action spaces.

Compare On-Policy Learning

On-Policy Learning vs Off Policy Learning→On-Policy Learning vs Q Learning→On-Policy Learning vs Deep Q Networks→

Learning Resources

📚
Reinforcement Learning: An Introduction - Chapter 6
book
→
📄
On-policy vs Off-policy Learning - OpenAI Spinning Up
docs
→
🎓
On-policy Methods in Reinforcement Learning - Coursera Course
course
→
🎬
On-policy Learning Explained - YouTube Video by DeepMind
video
→
📝
SARSA Algorithm Tutorial - GeeksforGeeks
tutorial
→

Related Tools

Reinforcement LearningSarsaPolicy GradientTemporal Difference LearningExploration Vs Exploitation

Alternatives to On-Policy Learning

Off Policy LearningQ LearningDeep Q Networks