Toyota Follows Tesla In Developing A Vision-Based Self-Driving System

image

According to a recent report published by Automotive News, Toyota's Woven Plant unit is working to develop advanced self-driving vehicle technology without the use of pricey sensors, including lidars.

Woven Planet is a subsidiary of Toyota that was formerly called the Toyota Research Institute – Advanced Development (TRI–AD). The subsidiary told Reuters it can collect the data used to train its self-driving technology by using inexpensive cameras. The goal here is to develop the system quickly and successfully without exorbitant costs. Vice president of Engineering at Woven Planet Michael Benisch shared:

"We need a lot of data. And it's not sufficient to just have a small amount of data that can be collected from a small fleet of very expensive autonomous vehicles."

Toyota acquired Lyft's self-driving division last year, and Benisch was the division's former engineering director. He added:

"Rather, we are trying to demonstrate that we can unlock the advantage that Toyota and a large automaker would have, which is access to a huge corpus of data, but with a much lower fidelity."

Toyota agrees with Tesla that it's imperative to compile a wealth of testing data from a large fleet of "self-driving" cars. It also agrees that using expensive sensors like radars and lidars in the testing process is expensive and not likely scalable.

When Tesla announced that it was switching to a camera-based vision-only approach for its advanced safety features and driver-assistance systems, there were worries that it wouldn't be robust enough. Tesla's vehicles even temporarily lost some of their impressive crash test ratings and recommendations. However, the automaker regained its status among safety organizations when it proved that the vision-only approach is working.

Tesla's approach allows it to put the hardware in all cars, which undergo testing in the real world by actual owners. All Tesla's vehicles come standard with Autopilot, though the brand's Full Self-Driving Beta technology is only able to be activated in vehicles owned and driven by Tesla-approved beta testers who paid for the feature and achieved a specific Tesla Safety Score over a period of time.

Meanwhile, companies like Alphabet's Waymo are still using expensive sensors and lidars in their vehicles, and they're testing in more closed environments, and/or with professional test drivers.

Woven Planet says its cameras are 90 percent less expensive than sensors it used in the past, and they're easy to install in fleets of passenger cars. Interestingly, the Toyota subsidiary notes that switching to the use of data from its cheap cameras has already increased system performance to that of when it was trained with only the data from expensive sensors.

image
image

Toyota did make it clear that it may still use advanced radars and lidars for robotaxis. Benisch explained that in many years it may be "entirely possible" for camera tech to exceed the performance of more advanced sensors. However, he still questions how long it may take to ensure a high level of safety and reliability.

Check out the related video from Automotive News TV below: