Servo - March 2008

Camera and sonar view of Pioneer 2.

world to model everything the robot perceives and responds to in the real world.

This way, the robot’s intelligence can attempt to sort of reason out what happens in the real world by use of the stored “working memory elements” of the virtual world, according to a Pace Lab Obstacle Avoidance paper, by Dr. D. Paul Benjamin, et al.

Machine Vision

The cognitive, machine vision robot — Pioneer 2 — sees through a pair of Fire Wire (IEEE 1394) Canon VC-C4 cameras and “framegrabbers.” The cameras are mobile via a pan/tilt hardware device from Directed Perceptions. The computer uses an onboard Linux OS for command and control, which interfaces wirelessly with a single PC-based computer.

The software brains of the Pioneer 2 consist of two parts: one that processes the “bottom-up” or simple view of the world; and, one that processes the “top down” or closer look view.

The bottom up view is quick and dirty, giving the background and the general lay of the land. This image simply provides a view with objects the robot may want to examine more closely. This software runs on Intel’s Open Vision software library.

The top down vision system is much more complex and elaborate. The robot’s system activates this element of the robot’s vision when the robot wants to take a closer look at something in the environment that the first vision system has only vaguely mapped out. This is for specific object recognition.

This “ERVision” software will look

Pioneer 2 with camera mount and cabling, top view. The P2 is the predecessor to the P3-AT from ActivMedia. This model has four wheels and motors with encoders. The AT model has optional sensing software to make utilize the new sensing hardware and turn that into intelligible commands the robot can follow.

The robot comes with upgrades including inertial correction that counters “skid steer dead reckoning errors.” The bot can be accessorized with laser range finders, both front and rear sonars, pan/tilt/zoom color cameras, stereo range finder cameras, and day/night vision cameras. Finally, GPS, color-tracking, compasses, and tilt position sensing hardware round out some of the most desireable add-ons. Internet operation is also available with this model.

at an object to recognize its distinct aspects. It will then model the image in the virtual world and store that so the object won’t have to be recognized again. This saves time and effort.

As the virtual world becomes more completely detailed, the robot’s memory and perception of its world becomes more clear and accurate. This way, the robot’s obstacle avoidance proficiency increases in its surroundings.

In case any of the obstacles should be mobile objects (say a house cat, for example, that might have been sleeping in one spot when it was originally recognized), the robot is able to determine whether any objects it has recognized before are now where they are supposed to be (whether they have moved) by use of the virtual world model.

SOAR

SOAR, an architecture for developing intelligent robot systems, has been in use for 25 years. It is

Pioneer 2 with a view of serial cable.

GEERHEAD

an integral part of the project that has produced the Pioneer 2 robot (shown in the images) with activevision.

The current version of the software is SOAR v8.6, for those who would like to investigate its use on their next project. SOAR developers hope to bring SOAR to the point where it can enable all the tasks of the kinds of intelligent robots the world envisions for the future.