一句话：提出模块化分布式架构，显式拆分 ML 流水线各阶段并服务化，使非专家学生一次工作坊内完成采集、训练与部署。
在 K-12 与入门教学中开展动手式机器学习教学面临多重困难：ML 流水线各阶段及其依赖关系不易被理解，快速实验与迭代难以实现，而教师往往缺乏 ML 与机器人方面的专业知识。已有平台要么对技术能力要求过高，要么把流水线隐藏在黑盒之中，限制了概念理解与实验学习。本文提出一种面向 ML 工作坊的模块化、分布式软件架构，以自动驾驶为应用场景，把流水线拆分为自包含阶段，并将硬件控制与模型执行封装为服务，使学生能在单次课程内完成数据采集、模型训练与真车部署。初步结果表明，中学生在三小时工作坊后对完整 ML 流水线及其因果关系形成了初步的概念性理解。
Machine learning (ML) literacy is increasingly seen as a core educational goal, yet hands-on ML teaching remains difficult in K-12 and introductory settings. Challenges include making ML pipeline stages and their interdependencies understandable, enabling rapid experimentation and iteration, and keeping workshops feasible for teachers without deep ML or robotics expertise. Many existing platforms either require substantial technical skills or hide the pipeline, limiting both conceptual understanding and experimental learning. We present a modular, distributed software architecture for ML workshops that makes all pipeline stages explicit and supports iterative experimentation, using autonomous driving as the application domain. The system splits the pipeline into self-contained stages and encapsulates hardware control and model execution as services. This allows students to collect data, train models, and deploy them to physical vehicles within a single session. Preliminary results indicate that secondary school students developed early conceptual understanding of the full ML pipeline and its cause-effect relationships in a three-hour workshop.
@inproceedings{paperbot11646,
title = {Modular Workshop Architecture for Teaching Machine Learning in Autonomous Driving by Non-Experts},
author = {Lukas Walter and Christopher Beck and Stefan Henkler},
booktitle = {ICSE},
year = {2026},
doi = {10.1145/3774748.3795673}
}