If you have ever driven a modern vehicle in a snowstorm, you know that the sensors on the car do not do well. As soon as the snow gets heavy, you can get warnings that your front collision warning is offline or some other safety feature is no longer working. The reason for this is simple; snow is complicated for sensors, especially as we develop autonomous vehicles.
However, an Ottawa-based software company is working on a solution by training artificial neural network models to make autonomous cars safer.
Sensor Cortek aims to develop a deep neural network and AI-based perception system that can handle almost anything mother nature can throw at a vehicle.
The company uses LiDAR sensors, radar, cameras and advanced GPS systems to log drives and the surrounding weather conditions. They complete these drives on a private 16-km track on its 1,850-acre site called Area XO. If Ottawa gets a storm, the Sensor Cortek team is out getting real-time data. With the sensors on their vehicles, the team can get about ten GB of data every minute of drive time.
After a drive, the data is uploaded to train artificial neural network models. Eventually, this artificial neural network can be paired with sensors on a vehicle to detect road users in poor conditions to allow an autonomous vehicle to make a safer decision.