Reinforcement Learning Training Framework Based on the Low Code and Behavior Tree

To simplify algorithm design and expedite algorithm training, this paper proposes a reinforcement learning training framework by integrating low-code development and behavior trees. Firstly, a low-code development framework is designed, enabling neural networks to be constructed and fine-tuned through drag-and-drop components, with training code automatically generated. Subsequently, behavior trees are constructed using expert knowledge, where behavior tree nodes represent agent actions, and prior knowledge constrains the exploration direction of algorithms. Experimental results demonstrate that this framework can streamline the algorithm construction process, optimize training modes, and shorten training cycles.