Train, test and validate Deep Learning Autonomous Driving
rFpro have released a short video showing their end-to-end simulation solution for testing, training and validating Deep Learning Autonomous Driving models.
A massively parallel test environment allows accurate, high quality, digital models of the real world to be shared by multiple ego vehicles, each with multiple sensor feeds, plus Swarm traffic and pedestrians, plus Programmed traffic and pedestrians, plus (when running real-time) Human test drivers in driving simulators.
The use of models of the real world is important. The high quality and stochastic nature of these models benefits the training performance of your deep neural networks.
Being able to add human test drivers into the scene is also very valuable for your training and testing data. Human drivers will add more unpredictable, slightly error-prone, behaviour per kilometer than any of your Swarm traffic.
The test environment can be shared over the cloud, with ego models, and human test drivers, joining from any of your locations worldwide. Invite other OEMs and Tier-1s to join the test environment so that different autonomous vehicles can test co-existence in simulation, while keeping your own IP completely protected.
Early 2018, rFpro will start hosting Open test environments in the cloud, for a variety of test locations across North America, Europe and Asia, for any subscribing OEMs or Tier-1s to join.