From an article in Mashable by Sasha Lekach
Why weigh down a self-driving car with a lot of sensors, HD maps, and equipment when you don’t have to?
That’s the philosophy of British startup Wayve. It claims it only needs a camera, GPS tracker, and a powerful computer to be able to drive anywhere autonomously.
But experts who specialize in sensing technologies like light-based LiDAR and radar say the idea mostly comes across as preposterous — or the very least, short-sighted.
Most self-driving cars decide how to drive down a street as it happens — picking up information about debris in the way, pedestrians on the sidewalk, the sun starting to set in the distance. Wayve doesn’t try to interpret that much data since it can’t really pick up much from its cameras. Instead it relies on training data already plugged into the system and information from past driving experiences, along with any “reinforcement” data, which is any information about the street from the human driver.
In a blog post about its human-driven method, the company said last week it knows its system takes a different and slower approach: “With each safety-driver intervention, our system learns and will improve, rather than buckle with scale. It will take us longer to reach our first deployment, but we are riding a fundamentally different curve.”
The experts Mashable talked to, however, didn’t think it was a great idea.
“It’s lunacy,” said Rick Tewell, COO at Velodyne LiDAR. “AI performs a lot better with a lot more data than less data.”
Self-driving cars ultimately need reliable, strong data sources to increase safety. Cost is certainly an issue, but as more of the sensing equipment is manufactured, the cost goes down. A LiDAR system used to be thousands of dollars only a few years ago. It’s already down to about $1,000 or less. Radar tech is even cheaper.
In terms of cost, Wayve says its sensor and computing costs are 10 percent of “traditional approaches.” But for car makers veering into autonomous vehicles and well-funded startups, saving some money for a simplified system isn’t a priority.
Leilei Shinohara, vice president of R&D at RoboSense, doesn’t understand why you wouldn’t want everything sensors can provide. “Sensors can get details that the human eye can’t,” he said in a phone call from China.
Even if you only have a LiDAR sensor as a safety redundancy, you want to have that on the vehicle — another fatal Uber or Tesla Autopilot crash has to be prevented. “You may only use LiDAR 5 percent of the time in certain situations if radar fails or is incomplete,” but you still want it for those instances, he explained. If a random scenario like a paint truck spilling paint all over a self-driving car happens, cameras and LiDAR sensors might be blocked, but radio-wave sensing radar system will take over.
For the full article click here.
