A virtual school for driverless cars
Waabi is a start-up that is disrupting the driverless car industry with an innovative new AI project.
The race to develop widely-available driverless cars continues, with companies and partnerships investing billions around the world in AI vehicles. Waabi is one start-up that is disrupting the driverless car industry, with a unique approach to testing autonomous vehicles.
The company is building Waabi World, a realistic virtual environment, where artificial intelligence “drivers” can be trained. The company aims for testing and training in the simulation, before practising in the real world.
Waabi was founded by Raquel Urtasun, who previously led the self-driving research team at Uber for four years. The CEO is also a Full Professor in the Department of Computer Science at the University of Toronto. “Right now most approaches to self-driving are just too slow to make progress”, she told Technology Review. “We need a radically different one”.
The virtual school for driverless cars
For artificial intelligence to understand the activity of everyday traffic, it must be trained to handle any event or experience. It has been reported that driverless car firms like Cruise or Waymo have invested in driving millions of miles around the world. “When you have an event that happens rarely, it takes thousands of road miles to test it properly”, said Sid Gandhi, a member of the simulation team at Cruise. “As we work on solving the long tail [events], we’ll rely less and less on real-world testing”.
Some companies have been able to test driverless cars without human drivers in calm urban environments around America. By subverting the cost of testing AI software in the real world, Urtasun intends to develop driverless cars at a faster pace, and cheaper too.
Simulation has become commonplace in the development of driverless cars. It has been reported that nearly all driverless car companies use the technology in some way. This is due to the fact that it is a quicker way to test, giving AI drivers a variety of events and scenarios that could be found on the road. Of course, it is less expensive too. That said, most driverless car companies currently mix simulated testing with real-world exercises. Waabi intends to focus solely on simulation.
Waabi World is an environment that is generated and operated using artificial intelligence. The technology can identify weaknesses in the AI driver, creating an environment that puts them to the test. The simulation can test multiple drivers simultaneously without any need for human input. As put by Technology Review, Waabi World makes AI a “driving instructor and stage manager”.
Pushing the limits of simulation
Waabi is perhaps the company that is most committed to using simulation to test driverless cars. Waabi World is based on data taken from real sensors on cameras, which is then used to create a replica of a real world environment. The AI driver reacts to this data, be it fog or light reflections, in a way that makes the virtual world even more realistic.
The driver understands how to navigate different scenarios, while another AI identifies weaknesses and tests them. By using one AI to test the other, more and more scenarios are explored, and eventually less weaknesses are found. “You will need to expose it to millions, perhaps billions, of scenarios in order to find your flaws”, said Urtasun. The CEO compares this to the way the human brain processes mistakes and learns new skills. “Every time that we experience something we rewire our brain”, he added.
With this new project, Urtasan aims to change the way driverless cars are tested, shortening the wait to ride them on roads around the world.