Erdos: Cost-Effective Peripheral Robotics for AI Education

Zachary Dodds, Ben TribelhornAAAI 2006

Abstract

The Erdos project addresses the "missing middle" in robotics education by transforming the iRobot Roomba into a sophisticated, sub-$200 research platform. By treating the robot as a peripheral device for an external laptop, the Erdos library provides a layered Python architecture that supports high-level AI algorithms. This approach democratizes access to embodied AI, allowing students to transition seamlessly from simulation to physical hardware.

Introduction

In the mid-2000s, robotics education faced a significant barrier. Research-grade robots like the Sony AIBO were prohibitively expensive for many undergraduate programs, while low-cost hobbyist kits often lacked the computational power and sensing capabilities required for modern AI curricula. The Erdos project was conceived to bridge this gap by leveraging the economies of scale of consumer electronics. By pairing a mass-produced vacuum cleaner with a student's laptop, the authors created a "peripheral robot" capable of running complex proofs-of-concept in localization and navigation.

System Architecture

The core innovation of the project is the Erdos library, a layered software framework designed to interface with the Roomba’s Serial Command Interface (SCI). The architecture is designed for modularity and cross-platform accessibility.

The layered software architecture of the Erdos library, showing the relationship between hardware, simulation, and high-level logic.

The library consists of several key components:

  • SRSerial: A simulation class that mimics the byte-level serial interface, enabling development without physical hardware.
  • Roomba Class: A low-level API for direct control of the robot's actuators (wheels, brushes) and access to its sensors (bumpers, cliff sensors, encoders).
  • Erdos Class: A high-level interpretive layer that handles complex logic, such as odometry integration and particle filtering.
  • Visualization Tools: Modules like Ervis and Erpyro provide geometric computation and integration with external frameworks like Pyrobot.

Methodology and Educational Deployment

The Erdos platform was evaluated through deployment in diverse educational settings, including introductory computer science courses and advanced robotics seminars. The "data" for these trials consisted of real-world and simulated environments, such as mazes used to test the Pledge algorithm and "Bug" navigation strategies.

A key feature of the library is its simulation-to-reality parity. The API is identical for both the simulator and the physical robot, allowing students to debug logic in a virtual environment before addressing the noise and unpredictability of physical sensors. This workflow was successfully implemented at Chatham College, where students progressed from learning basic functions in simulation to implementing complex loops and exception handling on physical Roombas.

Key Results and Insights

The Erdos project demonstrated that high-level AI research does not require high-cost hardware. The total system cost of approximately $175 made it significantly more accessible than contemporary alternatives.

The Erdos platform across various contexts: path planning simulations, physical hardware, and classroom deployment.

Key achievements include:

  • Advanced Algorithm Support: Despite its low cost, the platform successfully ran Monte Carlo Localization (MCL), using a 100-particle filter to track the robot's pose in real-time.
  • Robustness: By utilizing the Roomba, the project benefited from industrial-grade durability and integrated safety sensors (cliff and bump) that are often absent in custom-built educational kits.
  • The Peripheral Philosophy: Shifting the computational burden to an external laptop allows the robot to remain inexpensive while still benefiting from modern processing power.

Limitations and Future Directions

While successful, the initial Erdos platform had limitations. The base Roomba lacked long-range sensing (like sonar or IR ranging) and vision capabilities. Future work aimed to integrate webcams for vision-based SLAM and explore wireless tethering options to improve mobility. Despite these constraints, Erdos remains a landmark example of how consumer hardware can be repurposed to provide high-quality, low-cost education in embodied AI.

Summary updated

BibTeX
@inproceedings{dodds2006erdos,
  author = {Zachary Dodds and Ben Tribelhorn},
  title = {{Erdos: Cost-Effective Peripheral Robotics for AI Education}},
  booktitle = {AAAI Conference on Artificial Intelligence},
  year = {2006},
}
Figure