Toyota will develop autonomous driving systems that receive data from ordinary cameras rather than expensive sensors like lidars. The approach is similar to Tesla and explained by the trivial economy. The cameras are 90 percent cheaper than the sensors currently in use. Equipping a huge “fleet” of drones needed to train deep learning algorithms with them is easy – and it will not require serious expenses.
Since May 2021, the Tesla Model 3 and Model Y for the North American market are not equipped with radars. These are the first cars of the brand, whose Autopilot and Full-Self-Driving systems rely only on machine vision and a neural network. This approach is certainly beneficial because to create an adequately working autonomous driving system, you need to collect a huge array of data – and this can only be done with the help of millions of drones.
Otherwise, there simply won’t be enough data to train algorithms, says Michael Benish, vice president of engineering at Woven Planet. The company is a subsidiary of Toyota and is in the business of developing and implementing technology for drones.
By switching to cameras, the firm has saved up to 90 percent on each sensor, and the increased amount of data has boosted the system’s performance to the point where it would have trained only on lidars and other expensive sensors.
For now, all of these innovations are only for future developments. Toyota will still use lidar and radar on commercial robot taxis and other autonomous vehicles that will soon appear on the roads. The company is confident that today it is the safest solution. But in a few years, cameras may well supplant even the most advanced sensors.