Erdos: Cost-Effective Peripheral Robotics for AI Education

This work combines hardware, software, and curricula in order to create robots capable enough to advance the field of AI yet inexpensive enough to be widely accessible. Costs are kept low by pairing iRobot’s roombas with existing laptop or palmtop computers and their accessories. The result is a sub-$200 untethered physical platform capable of running and testing state-of-the-art AI algorithms. Motivation for Peripheral Robots The promise of peripheral robots is emerging atop two foundations: the powerful (and costly) selection of builtfor-research robots and the inexpensive (but far less capable) world of robotic kits. Figure 1 highlights the price/performance tradeoffs among several wellestablished and in-development robotic systems. Because iRobot’s roomba [1] vacuum offers a compelling balance among these factors, we at Harvey Mudd College are developing software, curricula, and a hardware interface atop this commercially successful base. We have named this freely available library of resources erdos for its potential to convert coffee into proofs: in particular, proofs-of-concept in which an embodied system might validate an AI algorithm or approach. Figure 2 contains a summary of the built-in capabilities [2] of the $150 RoombaRed platform. Figure 1’s $175 cost includes the USB-to-TTL converter required to interface with the robot's serial port. For $60 more, iRobot also sells a docking station which the robot can find and selfrecharge, provided a line-of-sight path to its IR emitter. Erdos: Software for the Roomba Figure 2 illustrates the components of the erdos library. The base operates as a serial device, either cabled or wireless, according to iRobot’s Serial Command Interface (SCI) specification [2]. SCI provides a byte-level interface; our SRSerial class simulates this interface byte-for-byte in case a physical platform is unavailable or unwanted. Copyright © 2006, American Association for Artificial Intelligence (www.aaai.org). All rights reserved. Figure 1 Comparing untethered robots, their prices, and their capabilities. Legend: nr indicates a platform not yet released as of 5/’06: prices are estimated. Computation: PC = personalcomputer-based, μP = microprocessor-based. Sensing: Vis = vision, Mic = microphone, Bmp = bump sensing, Cli = cliff sensing, Enc = encoders/odometry, Lt = light sensing, Son = sonar ranging, IR = infrared ranging, a2d = ttl-level inputs. Figure 2 At left is a block diagram of the erdos software library; at right, some of the roomba’s sensing and actuation abilities.