Angelus Oaks, CA--You can lead a robot to water, but how do you get him to avoid hitting the trees along the way? If you're Don Golding, president of Angelus Research, you install a new artificial-intelligence-based motion controller that mimics the logic processes of the human brain. Available in two versions--the ARC-110, and ARC-100--the controller endows robots or automated guided vehicles (AGVs) with the ability to execute high-level goals, as well as instinctive and behavioral responses for handling situations that are inherently unpredictable.
"We set out to give the machine enough common sense that it can work in an unstructured environment and react in real time to changes in that environment," Golding explains. Using the controller, AGVs can avoid people and temporary obstacles, and industrial robots--considered very dangerous around people--could be made much safer.
The controller accomplishes these objectives by imitating the brain's triune (three-level) architecture. It includes the electronic equivalents of instinctive, behavioral, and goal levels; a structure akin to the brain stem, limbic system, and cerebral cortex.
* Robotic search and rescue
* Nuclear power plant
* Bomb squad helper
Powered by a Motorola 68HC11 microprocessor, the controller accepts analog
inputs from infrared sensors. Custom sensors designed by Golding, these consist
of arrays of superbright LEDs alternated with optical detectors at Ĺ-inch
intervals. The controller pulses the LEDs rapidly and the detectors measure the
level of reflectance and compare it to the ambient light. Changes in reflectance
that exceed a threshold level--signaling an obstacle--trigger an
instinctive-level motor override.
Sensor arrays may be of almost unlimited length and might cover the sides or front of a robot or AGV. Though designed specifically to work with the analog output from the custom sensors, the controller can use a Motorola 68HC24 chip tied to the expansion port to accept as many as 16 additional inputs. Each card handles two axes of motion simultaneously. An on-board network port allows several controllers to operate together for more complex applications. Standard features include shunt-force feedback, advanced PWM speed/torque control, dual analog position inputs, four optical-sensor inputs, and one narrow-beam ultrasonic sonar subsystem (for navigation). Currently, the controllers only drive dc motors. But a modular board should be available in early 1996 that allows use of ac motors as well.
Users program goals, behaviors, and instincts on a PC using an English-like language. They then download commands to the controller via an RS-232 port and save it to 32k of on-board battery-backed RAM. The operating system and language are stored by the system in a separate 32k PROM.
As an example of the unique triune logic system's performance, consider the movement of a robot arm with sensor arrays lining the top, bottom, and left- and right sides of the arm. Programmed into the arm's controller would be a goal--for instance: "Move arm from Point A to Point B".
En route from A to B, should one of the sensor arrays detect an obstacle, an instinct-level command instantly overrides the task and performs real-time collision avoidance or causes the arm to stop. To extricate the arm from the problem, the controller then processes a prioritized behavior task list. Once the crisis has passed, the goal of moving to Point B resumes.
"Behaviors are an if-then structure," Golding says. "The programmer anticipates different situations the robot can get into, and creates behavioral responses for those situations." Currently, the controller is limited to 100 behaviors. But Angelus Research claims the typical application rarely needs more than a dozen.
The logic process is analogous to that executed by a person performing a simple task such as cooking. Should the chef accidentally touch a hot pot, he automatically pulls his hand away (instinct), runs cold water over the finger (behavior), and then resumes cooking (goal).
Goals, behaviors, and instincts can change on-the-fly to account for different situations. This capability can prove especially valuable for accomplishing multi-part tasks in which instincts useful in one task would prevent the successful completion of another. "While the robot searches for an object, you want collision avoidance," Golding explains, "but once it's found the object, you might then want the robot to move towards that object and not do collision avoidance."
He would love to take credit for intentionally aping the brain's structure, but the similarity is a serendipitous coincidence. "We were just solving an engineering problem," Golding says, "and this is what fell out."
In fact, early incarnations of Angelus Research's Intelligent Controller contained just two logic levels--goal and instinct. But without a behavior level, a robot or AGV could become stuck vacillating between two simultaneous and conflicting instinct triggers.
Golding believes his controller's unusual three-level architecture and simple operating system will allow one of his brainy little robots, the MR-1, to win an AAAI (American Association for Artificial Intelligence) navigation competition later this year. "Last year's winners had $20,000 worth of Sun workstations off-board the robot performing the navigation and collision avoidance processing, and they didn't actually solve the problem," he says. "Our machine is going to solve it automatically using two $400 controllers on-board."
Additional details...Contact Don Golding, Angelus Research, 6344 Sugar Pine Circle, Angelus Oaks, CA 92305-0098,
(909) 794-8325, http://www.angelusresearch.com .