While I attended the grand opening of the campus (that's still under construction), there are already employees at the site working. During a tour of one of the completed buildings, the automaker showed off how it plans to achieve autonomy in the future and change the way it works. Instead of a single team working on an entire system, sub-tasks (features of a system) are assigned to various departments to tackle problems in a fast-paced environment. BMW compared it to how startups work and touted how agile this will allow them to be.
The automaker noted that since deploying its test fleet in California, it has racked up 24 million kilometers (14.9 million miles) in data. But 95 percent of those miles were tested in a proprietary BMW simulator using data from its test cars, Pegasus the open-source scenario database, and a scenario editor. The result is something that looks like a video game where researchers fine-tune edge case situations to see how BMW's AI reacts to certain events.
Other companies like Mercedes, GM and Waymo also use simulators to test and train their AI. It's become an important part of the research. It's nearly impossible to encounter every situation while driving test vehicles. It's completely impossible to recreate and slightly tweak those situations to account for changing weather and road conditions and the presence of more (or fewer) vehicles, pedestrians and other objects.
Of course, all the car data has to come from somewhere. During a sensor demonstration, BMW shared a new solid-state LiDAR system that it says has double the range and 2.5 times the resolution of what's currently on the road today. On its test vehicles, laser-based sensors are placed in the front center and sides and rear. The result should be a higher-quality image of a car's surrounding area than what we've seen in the past.
Typically a sensor helps parse the objects it detects and passes it on to the vehicle's computer. BMW has decided to use the raw data directly from the cameras, LiDAR, radar, etc, combine it and do its own object detection. So instead of three pieces of hardware telling a computer if they see a person, the internal CPU uses the combined raw data to "see" a pedestrian.
BMW also believes mapping is part of that data-gathering package. "It's very important. We see it as a different sensor. With our onboard sensors, we are creating a live map of just what's happening around the car and we have a live map coming from Here," Dr. Wisselmann told Engadget. The live information and saved map data are overlaid to make sure the car is where it says it is.
Nokia's Here maps was acquired by BMW, Audi and Daimler in 2015. Mercedes uses it to for Active Distance Assist Distronic on the new S-Class to slow a car down if the map determines an upcoming corner is too sharp for the car's current speed.
BMW's iNext won't be the first level 3 car available for purchase. That distinction goes to Audi's 2019 A8. Unfortunately, there are regulatory issues that determine when and where the technology can be deployed. BMW is hoping by 2021 countries (and states) will have figured out how to accept vehicles like the iNext and A8.
Meanwhile, the automaker is determining how to actually navigate the autonomous future. The new Autonomous Driving Campus is not just a solid step in the right direction, it's a huge leap both financially and in how it builds cars generally. If an automaker isn't working on this issue, it risks being left behind.
BMW wants to make sure it does this right. It wants to be cautious. Seriously, they told me that over and over again. But that's important. Not just for the automaker but for the entire industry. Being careful about introducing these types of features keeps drivers safe and cars out of headlines.
BMW is known for fast cars. But fast doesn't necessarily mean unsafe and a giant facility dedicated to making sure its first self-driving cars are conservative on the roads is a good thing.
Source: Roberto Baldwin