This repository contains code and pre-trained models for a reinforcement learning approach to navigating microrobotic swarms in unknown environments with dynamic obstacles. It implements a Proximal Policy Optimization (PPO) algorithm with a Transformer-based policy network, training agents to autonomously explore and avoid both static and moving obstacles. The system includes a training pipeline with parallel workers, configuration files, and deployment code for real-world control using a three-axis Helmholtz coil system and YOLOv5 object detection. The project targets microrobotics applications requiring intelligent swarm navigation under uncertainty.