The AutoDrive Challenge is a 4+ year competition sponsored by SAE and General Motors to develop and demonstrate a fully autonomous driving passenger vehicle. Each year of the competition there will be new challenges building towards the goal of navigating an urban driving course. Currently, there are at least 8 universities competing against each other to create the best looking, best designed, and best functioning vehicle.
At MTU, the goals of the competition are split between several teams. Each team creates a solution for their specific goal, and then all of the solutions are brought together to create a unique autonomous vehicle that we can all be proud of! The teams include: the Me-em Team, the Mathworks Team, the Mapping Team, and the Perception Teams.
The Me-em Team designs and builds the mounts and other physical connectors on the autonomous car. With any autonomous car, there are always sensors that scan and analyze the world around us. Each of these senors require some sort of stable, strong, and well designed mount on the car. Furthermore, the Me-em Team will often research what types of sensors are needed for the competition, order the chosen sensors, and implement them into the car.
In the past, they have worked with LIDAR and camera sensors, and are currently looking into RADAR sensors and how to mount them to the car.
Simulation is critical in the design of a car because it allows engineers to find flaws in their designs and code before actually building and testing an expensive real life car. In the Mathworks portion of the AutoDrive Competition, the competition requires the use of Mathworks to simulate various tasks the car might perform.
Last year the car was required to run through a simulation where it drove autonomously through a simulation and avoided obstacles. This year, the Mathworks Team works to refine the simulation by designing a path-finding and steering method based off of sensor data. Improvements here, can greatly improve the performance of the autonomous car in the competition.
Once an autonomous car is built and functioning, it needs to be able to make its way to some destination, and without a mapping/GPS program, it can’t go anywhere. In the Mapping portion of the competition, the car is required to be able to route from its current location, to some destination on the track.
Built from the ground up, the mapping program can find GPS points, addresses, and locations on a map, and can even route between two points. Currently, the Mapping Team is optimizing code within the program and adding the ability reroute around obstacles on the road. In the future, the mapping program will be loaded on the autonomous car to be used in the competition.
All autonomous cars depend on GPS and various other measures to find their exact location on a map. Localization is the use of algorithms and sensor data to estimate the vehicles location. Once Localization methods are combined with GPS, the internal computer can have an exceptionally accurate location on the map.
Currently, the Localization Team is in the process of researching methods to evaluate Localization. These methods will be used to improve the vehicles Localization accuracy in the future. They’re also working on their section for the Concept Design Report. In this section, they have to explain how they will deal with GPS dropouts while maintaining Level 4 Autonomy.
While most people in cars can easily identify a person crossing the street or a stop sign just ahead, it’s not as easy for a car, and if a company is planning on releasing an autonomous car, it better be able to recognize people and shapes around it.
To improve stop sign, stop light, and people detection, the Object Detection Team has invested time into training a neural network with countless images from a massive database of marked images. With the improved recognition a properly trained network brings, the autonomous car will be able to more easily recognize signs and obstacles on the competition track.