Abstract
Robotics is a rapidly expanding field that needs software engineers. Most of our undergraduates, however, are not equipped to manage the unique challenges associated with the development of software for modern robots. In this work we introduce a course we have designed and delivered to better prepare students to develop software for robot systems. The course is unique in that: 1) it emphasizes the distinctive challenges of software development for robots paired with the software engineering techniques that may help manage those challenges, 2) it provides many opportunities for experiential learning across the robotics and software engineering interface, and 3) it lowers the barriers for learning how to build such systems. We describe the principles and innovations of the course, its content and delivery, and finish with the lessons we have learned.
Introduction
While the robotics industry is experiencing unprecedented growth, a significant talent gap exists. Traditional computer science (CS) undergraduates often struggle with the "messiness" of the physical world. Standard software engineering (SE) curricula typically focus on deterministic virtual environments where inputs are clean and predictable. In contrast, robotics involves "leaky abstractions"—where hardware limitations frequently break software assumptions—and inherent uncertainty from sensors and actuators.
To address this, we developed a novel undergraduate course at the University of Virginia designed to bridge the gap between traditional SE and robotics. The course aims to move beyond narrow hardware-centric mechatronics or high-level AI theory, focusing instead on the software engineering rigor required to build maintainable, reliable robot systems.
Pedagogical Principles and Methodology
The course is built upon five foundational principles designed to prioritize the unique intersection of robotics and software engineering.
Central to our methodology is the "Pairing" Strategy. Every robotics concept is taught alongside a corresponding SE tool or practice. For instance, when students learn about robot state, they are introduced to Finite State Machines (FSMs). When discussing sensor noise, the curriculum emphasizes logging, unit testing, and data filtering. This approach re-contextualizes SE principles like modularization and component reuse as essential solutions to physical robotic problems rather than abstract academic exercises.
Course Implementation and Simulation
To lower the "hardware tax"—the high cost and fragility of physical robots that often bar entry for students—we developed a custom, lightweight drone simulator based on FlightGoggles. This simulator is computationally efficient enough to run on a standard laptop within a Virtual Machine (VM), ensuring a consistent, zero-cost development environment for all students.
The course utilizes the industry-standard Robot Operating System (ROS) as the primary framework. Students progress through incremental abstractions, starting with high-level "black box" commands (e.g., "go to waypoint") and gradually peeling back layers to implement low-level PID controllers and perception algorithms. This progression follows the Sense-Compute-Act cycle, forcing students to account for the environment's impact on their software logic.
Key Insights and Results
Our findings suggest that simulation is a feature, not just a cost-saving measure. By "injecting" specific amounts of noise and uncertainty into virtual sensors, we forced students to write more robust code than they might have in a controlled lab with perfect hardware.
Furthermore, the "checkpoint" system—verbal demonstrations with a 1:5 teacher-to-student ratio—proved vital. These sessions required students to explain their design decisions and reflect on how their software handled physical edge cases, preventing them from simply "hacking" a solution that worked once by chance.
Key outcomes include: * Accessibility: The VM-based simulator successfully ran on systems with as little as 2GB of RAM. * Industry Impact: Approximately 10% of students from the initial cohorts have already entered the robotics industry. * Resilience: The course successfully transitioned to a fully virtual format during the COVID-19 pandemic without losing its hands-on pedagogical value.
Conclusion and Future Work
While successful, we identified areas for improvement, such as the "time tax" experienced by students on older hardware during intensive image processing tasks. Future iterations will migrate from ROS 1 to ROS 2.0 to align with industry shifts and will incorporate more formal software specifications into the planning phases of the labs. Ultimately, we aim to develop formal instruments to empirically measure the performance of these students compared to those in traditional CS tracks.
Summary updated