Teaching Artificial Intelligence and Robotics Via Games

Daniel Wong, Ryan Zink, Sven KoenigEAAI 2010 · DOI

TL;DR

This paper introduces two innovative projects designed to teach core Artificial Intelligence (AI) and robotics concepts to computer science students using video games as a motivational tool. By developing standalone AI projects that leverage game mechanics and creating a hardware/software interface for controlling physical pinball machines, the authors aim to increase student engagement, retention, and practical skill development in these fields.

What Problem Does This Paper Address?

The authors identify a challenge in traditional computer science education: a potential lack of student engagement and motivation, particularly in foundational areas like AI and robotics. They observe that while new degree programs in game development are emerging, there's a need to integrate engaging, game-centric approaches into broader computer science curricula. Specifically, they aim to bridge the gap between theoretical AI/robotics concepts and their practical application, making these subjects more accessible and appealing to students, including those not specializing in game development. The paper also highlights the disconnect between software-focused computer science education and the hardware-interfacing skills crucial for robotics.

Methods

The paper presents two distinct projects:

  • Computer Games in the Classroom Project: This project focuses on developing standalone AI modules that use video game scenarios as a motivational context. These modules do not require students to use complex game engines, making them accessible to a wider range of computer science students. The projects cover standard AI algorithms such as:
  • Fast Trajectory Replanning with Variants of A*: Students implement and extend the A* search algorithm and the recently developed Adaptive A* for pathfinding in game environments.
  • Any-Angle Path Planning with Variants of A*: Students implement and extend A* to the any-angle search algorithm Theta*, focusing on path planning in gridworlds.
  • Gesture Recognition with Neural Networks: Students use neural networks and the back-propagation algorithm to recognize hand gestures for video games, extending existing machine learning concepts.
  • Pinball Project: This project addresses the gap in teaching hardware-software integration for robotics. It involves developing a PC interface for a physical pinball machine, allowing students to learn about actuators, sensors, signal generation, and real-time programming. Students design and implement a pinball game ("Pinhorse") on this platform, demonstrating control of a physical system.

Datasets

Not discussed in the paper. The projects described are educational tools and do not appear to rely on pre-existing, large-scale datasets for training or evaluation in the traditional sense. Instead, they utilize game scenarios and physical hardware as the "environment" for learning.

Key Insights and Ideas

  • Games as Powerful Motivators: The core idea is that video games can significantly boost student interest, motivation, and retention in traditionally challenging computer science subjects like AI and robotics.
  • Bridging Theory and Practice: By framing AI algorithms within game contexts (pathfinding, gesture recognition) and by interfacing with physical robots (pinball machines), students gain a more tangible understanding of theoretical concepts.
  • Accessibility of Game-Based Learning: The "Computer Games in the Classroom" project successfully decouples game-based learning from complex game engines, making it accessible to students who may not be game development specialists.
  • Demystifying Robotics Hardware: The Pinball Project offers a practical, engaging entry point into robotics by using a familiar, albeit mechanical, system to teach fundamental hardware-software interaction principles.
  • Future Potential for Adaptive AI: The authors hint at integrating AI techniques into pinball games to adapt difficulty, suggesting a future direction for personalized learning experiences.

Key Results

The paper focuses on the pedagogical approach and the development of educational projects rather than presenting quantitative performance metrics. The "results" are qualitative:

  • The projects have been successfully used in AI and introductory robotics classes at the University of Southern California and have been shared with other universities.
  • The Pinball Project successfully demonstrated control of a physical pinball machine using a PC interface, a feat the authors claim is novel.
  • The projects are designed to teach specific AI algorithms (A*, Adaptive A*, Theta*, neural networks) and robotics concepts (actuators, sensors, hardware interfacing).

Limitations and Future Work

  • Evaluation of Effectiveness: The authors explicitly state that it is "future work to evaluate the effects of our efforts," meaning the long-term impact on student learning and retention has not yet been rigorously measured.
  • Scalability of Gesture Recognition: While neural networks are used for gesture recognition, the paper notes they are not ideal for reliably classifying a large number of gestures without preprocessing, suggesting more specialized techniques might be needed for complex scenarios.
  • Future Work: The authors plan to explore integrating AI techniques into pinball games to adapt difficulty for players. They also mention making their pinball interface available for further research and teaching.

Summary updated

BibTeX
@inproceedings{wong2010teaching,
  author = {Daniel Wong and Ryan Zink and Sven Koenig},
  title = {{Teaching Artificial Intelligence and Robotics Via Games}},
  booktitle = {EAAI},
  year = {2010},
  doi = {10.1609/aaai.v24i3.18836},
}
Figure