Autonomous research robot
Autonomous research robots are a relatively new phenomenon. In the 1980s and even the 1990s, most researchers built their own platforms from scratch, so research proceeded at a snail's pace.
Pioneers
The Denning Mobile Robot company of Boston was the first to offer ready-made autonomous robots, which were purchased primarily by researchers. Grinnell More's Real World Interface, Inc. (RWI) and James Slater's Nomadic Technologies in the US and Francesco Mondada's K-Team in Switzerland were also among the pioneers to address the need for ready-made robots for robotics researchers with the B-21 from RWI, XR4000 from Nomadic and tiny Khepera mobile robot from K-Team; however, prices meant that only a few graduate students and military researchers could afford them. The low-cost Pioneer robot was introduced by a collaboration between RWI and ActivMedia Robotics in 1995, making robots available to many and opening the floodgates to new research in mobile robotics.
By 1999, the Denning company was defunct. In 1998, RWI joined with ISRobotics to form the iRobot corporation. There Grinnell More introduced the PackBot remote control robot, veering away from autonomous development and research robots to pursue military venues. Nomadic Technologies also left the field. MobileRobots Inc and K-Team continued to build on autonomy and provide for the research community.
In 2003 the Defense Advanced Research Projects Agency (DARPA) contracted with Segway to convert fifteen Segway PTs into Segway Robotic Mobility Platforms. Segway developed the platform to serve as a reliable, cost-effective tool for research institutions and delivered the units to DARPA in April. In June 2003 DARPA worked with SPAWAR Systems Center, San Diego to distribute the units to 14 government and university research institutions to use in robotic research projects.[1].
Research robots made great strides in autonomous indoor navigation between the 1990s and the 2000s. Currently, a number of ready-made research bases have the sensing, mobility, and computational power necessary for such autonomy: The Pioneer, PatrolBot, PowerBot and PeopleBot platforms can map buildings and navigate out-of-the box, using SLAM and a variation on Monte Carlo method/Markov localization and modified value-iterated search navigation techniques, with any sensor of the 2-D range-finder class. This method creates a human readable map of the robot's workspace that can be used to control and track robots of this type as they move. Evolution Robotics offers single-camera VSLAM software, which replaces range-finding with visual pattern-matching, but this system cannot create a human readable map with which to monitor robots' position. Other groups are building stereocam-based VSLAM. Because the stereocam provides range-finding data using the disparity between the lenses, maps can be made and robots tracked. The K-Team Khepera, Segway-based platforms and other research robots can link to external computing resources to use such software.
The precision of any of these methods depends upon the precision of the sensor, the granularity of data tracked and the speed of calculation. Range-finding lasers may have +/-1 cm accuracy while digital stereocamera accuracy is limited to a quarter pixel and thus is highly range-dependent. Vision-based systems require more computational resources than simple range-finding systems such as lasers, but may do the computation on a digital signal processor embedded with the camera. Because of cost and precision trade-offs, less expensive vision-based systems tend to be used on consumer robots while commercial and industrial robots and automated guided vehicles (AGVs) tend to use laser-based systems.
Outdoors, localization is primarily handled with GPS, however, satellite signals can frequently be lost due to weather, trees, buildings or other obstructions. When the signal is lost, the robot typically navigates using dead reckoning and inertial motion tracking. Dead reckoning relies on relative wheel motion and is highly subject to cumulative slippage errors. Inertial motion tracking uses rate gyroscopes and accelerometers to determine actual motion of the platform. The accuracy of inertial motion tracking depends upon the quality and calibration of the sensors employed. The Segway RMP 400 and Seekur robots are two of the few research platforms designed for such research; most other outdoor research robots are jerry-rigged by researchers from existing vehicles.
In constrained areas, some robots, such as the John Deere Gator, simply surround the perimeter with radio beacons and use simple triangulation from three or more beacons to localize and navigate. Beacons are also used indoors by older AGVs in factories.
Autonomous Solutions is a leader in the field of outdoor navigation software; their system is used by John Deere tractors and by some military platforms.
Programming research robots
Much research software for autonomous robots is Free Software or Open Source Software, including Carmen from Carnegie Mellon, Player/Stage/Gazebo from the University of Southern California and the ARIA API libraries from MobileRobots, Inc. There is also commercial software: Webots has been continuously developed since 1998 and is currently used by more than 500 universities. It runs on Linux, Windows and Mac OS X. More recently, in June 2006, Microsoft Research began offering free beta-test copies of a Robotics Studio software development kit with Pioneer robots in simulation for Windows XP in an attempt to counter Linux dominance onboard mobile robot platforms. An older platform: URBI with a Free Software SDK is used in many universities. The plethora of autonomous mobile robots and software available for researchers has greatly sped the pace of development in the robotics field.
