research Flow Models Exploration Conditional image generation with flow matching, implemented from scratch on MNIST and CIFAR-10. MeanFlow for Image Inverse Problems One-step image denoising by starting MeanFlow from noisy images instead of Gaussian noise. GMFlow for Inverse Problems Variational and Gaussian-mixture flow matching for image inverse problems, validated on 2D toy examples. Deep Reinforcement Learning Self-study of UC Berkeley CS285, with core deep RL algorithms implemented from scratch. engineering RoboMaster Hero Robot Real-time electrical control system on an STM32H723, with quaternion-based IMU orientation estimation.