From Google to Tesla, companies around the world are working on developing driverless cars. Manufacturers have had various degrees of success thus far, but the overall progress has been good, and many experts predict driverless cars to be widespread on the road within the next decade. One car manufacturer is taking matters into its own hands to develop the best driverless technology on the market—here’s how Toyota is teaching its cars to drive.

The initiative started in January 2016 with the creation of the Toyota Research Institute (TRI), the company’s $1 billion investment in developing artificial intelligence for driverless cars and home-care robots. Because the institute is still relatively new, its main focus is currently on research and development; a large push right now is to collect a variety of data from vehicles, including from sensors, cameras, radars, and LIDAR. The eventual goal is to create effective driverless technology that will work in all Toyotas, as well as other brands of cars.

Once it fully understands the data from the car, TRI will combine it with data from outside the vehicle, including mapping roads and traffic patterns. The goal is to eventually get driverless technology to the same level of humans where it doesn’t necessarily rely completely on maps but instead uses perception to see where it is and where the obstacles and difficulties lie ahead, but that likely won’t occur until sensor systems are stronger and more reliable.

One of TRI’s more unique tactics is to learn from drivers and apply that data and learning to driverless technology. Machines are trained by example rather than by rules, so safe driverless cars need to learn from examples, such as human drivers. Toyota does this in part by having humans label various driving videos feeds to identify safety threats such as a bike in the road, an oncoming car, or a pedestrian. It can then take that human-created data and run it through a machine with numerous other scenarios so the technology can adopt those human-like obstacle identification skills. The end result is a model that can identify objects on a video in real time, very similar to how a human would do it. Because Toyota is an international company that has cars all over the world, it also needs to be able to create driverless models that fit in anywhere in the world. How you drive in Manhattan is different than how you drive in the suburbs of Minneapolis, so the model takes into account human driver characteristics and trends to create profiles of sorts for various locations.

Like most other car companies, Toyota already sells cars with driverless features, including driver-assistance technologies like lane assist and rear-view cameras that are available in many cars including the Toyota RAV4 Hybrid. According to leaders at TRI, cars will likely continue to get safer and more automated incrementally until fully autonomous cars are ready to hit the open road, with autonomous features starting with the most dangerous areas and working forward. It is much more likely that safety regulators and human drivers will accept a car with multiple driver-assistance technologies that are proven safe over a fully driverless car that isn’t as proven.

TRI is definitely making progress in the driverless space, but the work isn’t without its challenges. Dealing with such large amounts of data from car sensors, road maps, and driver research can be an incredibly large undertaking, especially when the data is constantly being updated and upgraded with new technology.

What’s next for driverless technology at Toyota? Continued research and a huge push towards simulation. In order to prove safety and continue development, the company will have to spend hundreds of hours and millions of miles testing its technology. Simulations allow researchers to isolate various dangerous conditions and make sure the car performs well in those scenarios, instead of simply testing the car on hundreds of miles of flat roads on sunny days. The company also plans to continue developing its technology as new research and data becomes available. Because driverless technology is constantly changing, the work is never really done.

We could be riding around in driverless cars before we know it. And thanks to the work being done at TRI, there’s a greater chance that those driverless cars just might be Toyote
Technology
 Rick Delgado Rick Delgad