Shallow Blue: Lego-Based Embodied AI as a Platform for Cross-Curricular Project Based Learning

Robert Selkowitz, Debra T. BurhansAAAI 2014 · DOI

TL;DR

This paper introduces "Shallow Blue" (SB), a low-cost, Lego-based embodied AI chess player designed for interdisciplinary undergraduate learning. SB demonstrates how students can engage in complex hardware and software development, fostering skills crucial for careers in AI and robotics, even at smaller institutions.

What Problem Does This Paper Address?

The authors address the challenge of providing meaningful, hands-on AI and robotics education, particularly at smaller undergraduate institutions with limited resources and faculty. Traditional approaches often rely on pre-built platforms or require expensive custom hardware, limiting student involvement in core design and problem-solving. The paper highlights the need for educational tools that bridge the gap between theoretical AI concepts and real-world engineering challenges, enabling students to tackle complex, multi-faceted projects.

Methods

The core of the project is "Shallow Blue" (SB), an embodied AI chess player built using Lego Mindstorms. The system comprises five coordinated subsystems:

  • Physical Infrastructure (Hardware): A Cartesian crawler design using Lego components for movement across the chessboard. This includes a bridge and rail system for translation and a tower with a gripping claw for piece manipulation.
  • Master Control Program (MCP): The central software orchestrating the entire system. It handles communication between subsystems, issues motion commands, queries vision for board state updates, and interacts with the chess engine.
  • Image Processing System (Vision): Written in C++ and utilizing OpenCV, this subsystem processes board images to detect piece positions and analyze opponent moves.
  • Motion Controllers (Movement): Responsible for executing the physical movements of the robot based on commands from the MCP.
  • Chess Decision Making (Houdini): An external chess engine that provides the robot's moves. The system uses a custom interface to communicate with Houdini.

A key methodological innovation is the iterative hardware design process, particularly for the claw mechanism. Students went through multiple prototypes, learning to balance design constraints like space, functionality, and grip strength. The paper also emphasizes the hybrid hardware/software approach to problem-solving, exemplified by the oscillation issue, where both mechanical adjustments and software control were necessary for a solution.

Datasets

Not discussed in the paper. The system is designed to play chess against human opponents or other chess engines, and its performance is evaluated through gameplay and participation in competitions like the AAAI Robotics Fair.

Key Insights and Ideas

  • Lego as a Powerful Educational Platform: Contrary to some views, Lego Mindstorms is presented as a sufficiently flexible and cost-effective platform for deep undergraduate AI and robotics education, allowing for significant hardware and software design.
  • Cross-Curricular Synergy: The project thrives on interdisciplinary collaboration between computer science, physics, and engineering students, mirroring real-world R&D environments.
  • Embodied AI for Enhanced Learning: The physical embodiment of the AI agent (playing chess) creates a more engaging and intuitive learning experience, fostering a sense of connection and humanization that goes beyond purely software-based AI.
  • Iterative Design and Trade-offs: Students learn valuable lessons about the engineering design process, including the necessity of iterative refinement and making trade-offs between competing design goals (e.g., centering capability vs. grip strength).
  • Bridging the Gap to Professional Practice: The project provides students with experience in areas often not covered in typical CS curricula, such as managing multi-device communication, integrating diverse software components, and dealing with real-world constraints like noise and hardware limitations.

Key Results

  • Successful Embodied AI Agent: Shallow Blue was successfully developed and demonstrated at the AAAI Robotics Fair, performing reliably in chess matches.
  • Student Skill Development: Participants reported substantial development in skills applicable to post-undergraduate education and careers.
  • Tangible Outcomes: Student outcomes included senior theses, conference presentations, peer-reviewed publications, and admission to graduate programs.
  • Cost-Effectiveness: The entire system, including vision, had a total assembly cost below $800, making it accessible for undergraduate projects.
  • Improved Learning Experience: Students reported a sense of connection to the robot, and playing against SB was described as a more engaging experience than playing against a purely onscreen computer opponent.

Limitations and Future Work

  • Hardware Limitations: Lego's structural and motor limitations, as well as the inherent trade-offs in actuation and structural elements, are acknowledged.
  • Vision System Sensitivity: The initial vision system struggled under dim lighting conditions, requiring an upgrade with a studio light system.
  • Future Work:
    • Upgrading the vision system for mid-board game state detection to allow SB to start from mid-game positions or solve chess puzzles.
  • Exploring the human-robot interaction aspect of playing against SB, in collaboration with social scientists, to understand the phenomenon of humanization.
  • Extending the collaboration beyond traditional STEM fields.
  • Developing new AI systems using the Lego platform.

Summary updated

BibTeX
@inproceedings{selkowitz2014shallow,
  author = {Robert Selkowitz and Debra T. Burhans},
  title = {{Shallow Blue: Lego-Based Embodied AI as a Platform for Cross-Curricular Project Based Learning}},
  booktitle = {AAAI Conference on Artificial Intelligence},
  year = {2014},
  doi = {10.1609/aaai.v28i3.19040},
}
Figure