[Paper Review] Software Testing, AI and Robotics (STAIR) Learning Lab

Simon Haller-Seeber, Thomas Gatterer+4 more|arXiv (Cornell University)|Apr 6, 2022

Robotics and Automated Systems4 citations

TL;DR

The STAIR Learning Lab introduces a dual physical-simulated robotics platform integrating AI, software testing, and educational robotics for K-12 STEM learning. Using a custom Sphero RVR and Lynxmotion arm with ROS, it enables hands-on AI training (e.g., traffic sign recognition) via Teachable Machine, SVM, and YOLO models, with seamless testing across digital twins and real hardware.

ABSTRACT

In this paper we presented the Software Testing, AI and Robotics (STAIR) Learning Lab. STAIR is an initiative started at the University of Innsbruck to bring robotics, Artificial Intelligence (AI) and software testing into schools. In the lab physical and virtual learning units are developed in parallel and in sync with each other. Its core learning approach is based the develop of both a physical and simulated robotics environment. In both environments AI scenarios (like traffic sign recognition) are deployed and tested. We present and focus on our newly designed MiniBot that are both built on hardware which was designed for educational and research purposes as well as the simulation environment. Additionally, we describe first learning design concepts and a showcase scenario (i.e., AI-based traffic sign recognition) with different exercises which can easily be extended.

Motivation & Objective

• To bridge the gap in educational robotics and AI integration in Tyrolean schools by creating accessible, hands-on learning experiences.
• To develop a low-cost, robust, and extensible robotics platform combining physical computing with simulated environments for teaching AI and software testing.
• To support diverse learning levels through tiered exercises—from block-based programming to advanced ML model deployment—on both real and virtual robots.
• To establish a standardized ROS-based software stack for the Sphero RVR and Lynxmotion arm to enable reproducible, testable, and extensible educational workflows.
• To evaluate and improve learning outcomes through evidence-based course development and teacher-informed material design.

Proposed method

• Designed a hybrid robotics platform using Sphero RVR as a mobile base, Lynxmotion 4-DoF arm for manipulation, and Nvidia Jetson Nano for on-board AI processing.
• Integrated Intel RealSense D435 depth camera for 3D perception and sensor fusion, enabling advanced AI tasks like object detection and SLAM.
• Built a standardized ROS middleware stack to unify control, communication, and testing across both physical and simulated environments.
• Enabled cross-environment execution using ROS plugins for Visual Studio Code and IntelliJ IDEA, supporting automated testing with multiple test frameworks.
• Developed a multi-tiered learning scenario around AI-based traffic sign recognition, using Teachable Machine (beginner), SVM (intermediate), and YOLO (advanced) models.
• Provided a digital twin simulation environment that mirrors the physical robot’s behavior, allowing identical code execution and testing in both domains.

Experimental results

Research questions

• RQ1 How can a unified physical-simulated robotics platform support scalable, hands-on AI and software testing education in K-12 schools?
• RQ2 What are the most effective tiered learning pathways for introducing AI concepts—from basic model training to advanced deep learning—using accessible hardware?
• RQ3 How can ROS-based middleware enable consistent, testable, and extensible software development across physical and simulated robotics environments?
• RQ4 To what extent can low-cost, open-source robotics components support advanced AI and software testing tasks in educational settings?
• RQ5 How can teacher-informed learning design and evidence-based evaluation improve the effectiveness of robotics education programs?

Key findings

• The STAIR Learning Lab successfully developed a fully functional, ROS-integrated robotics platform combining Sphero RVR, Lynxmotion arm, and Jetson Nano, enabling real-time AI inference and software testing.
• The platform supports end-to-end AI workflows—from data collection and model training in Teachable Machine to deployment and testing on both physical and simulated robots.
• The use of a digital twin allows identical code execution and testing across environments, significantly improving software quality and reducing debugging time.
• The tiered learning approach enables students from beginner to advanced levels to engage with AI, from block-based programming to training and evaluating YOLO models.
• The ROS-based software stack, enhanced with VS Code and IntelliJ IDEA plugins, supports automated testing and version control, ensuring robust software development practices.
• Preliminary testing with GMAR summer school participants validated the platform’s usability and identified key software improvements, supporting future evidence-based course development.