Team Gray may not have advanced to the finals but they’ve got their sights on bigger targets. With their latest autonomy in-a-box they may be positioned to become the Microsoft and IBM of autonomous vehicles.
Chief Engineer Paul Trepagnier explained that the Autonomous Vehicle System, or AVS for short, is a completely self-contained unit comprised of componentized navigational mobility that can be scripted.
The system effectively boils the entire autonomous navigation down into three major categories:
- Localization
- Obstacle Avoidance
- Actuation
It takes input from a GPS sensor to determine where it is in the world and the desired route to traverse. “The core AVS platform takes the GPS information and drives the vehicle. We’ve driven the car up to eighty miles per hour with less than 5 cm of error,” Trepagnier said. “Eighty two miles per hour,” President and Director of Gray Matter Eric Gray corrected.
The system is designed to accept input from sensors in order to identify obstacles. In the event an obstacle is in the way of the vehicle, the AVS performs the appropriate behavior to avoid the obstacle if possible.
As a testament to the modularity and flexibility of the system, Trepagnier explained that when they acquired the new vehicle for the National Qualifying Event, retiring “the white one,” they were able to integrate all sensors and actuators in less than fifteen minutes.
The modularity sheds a little light into the elegant architecture in the design. First, the system is comprised of two major layers: The Hardware Layer and The Software Layer. Both are proprietary.
Features of The Software Layer include the extensibility of the sensor array. It takes only three days or less to integrate a new sensor. The integration is comprised of a reusable driver that provides a common output that is consumable by the obstacle detection algorithms. Considering the available-now and relatively inexpensive cost of this device it behooves all major sensor manufacturers to develop AVS-compliant sensor drivers for their technologies.
Also within The Software Layer is a high-priority fail-safe system. There are two threads constantly running that monitor the software and the sensors: a Safety Monitor and a Failure Monitor, respectively. If one of the threads identifies something wrong, they have the authority to bring the vehicle to a stop.
The Hardware Layer also has some important responsibilities. “We didn’t like that (the monitor threads) could potentially go down. So we designed hardware to monitor the Safety Monitor and the Failure Monitor. The Hardware Layer can stop the vehicle if it detects that something went wrong with the monitors.” The Hardware Layer is connected directly into the vehicle’s drive-by-wire system. The system has no moving parts and can withstand “way above 10g’s. Most of what we’ve seen in our applications have incurred fewer than 2g’s.”
The team has several prospects with whom demonstrations and negotiations are underway. The base price of the unit is $125,000. This is a tremendously cost-efficient solution to an otherwise expensive problem.
“Our vision is to be a commonplace technology in all vehicular testing environments,” declared Eric Gray. “Universities are talking about using this as a basis for their work. It allows them to focus their efforts on areas of specialization such as computer vision.”
Trepagnier seemed to enjoy his next hypothetical scenario almost a little too much. “We have a computer that can drive within five centimeters of a planned course. We can run the car in record mode, drive the track, and then put the path through a post-processor that will run a genetic algorithm to find the optimal route, and feed it into AVS. ” He said of racing an autonomous vehicle at high speeds. “We would be unbeatable.”
