Checklist for winter-proof autonomous car development
My own 6-year old Sedan is far from autonomous, but it has a parking sensor. This winter, the parking sensor was not functioning for at least 5 weeks in a row, which means I've been re-practicing reversing the car in full manual mode, like in the good old days.
In winter, car controls remain the same, but the driving AI must consider many factors impacting driving experience and safety:
I visited the Consumer Electronics Show (CES) 2019 in Las Vegas; an event which has grown into one of the leading automotive technology events. Vendors were showcasing their core autonomous driving solutions and roadmaps: Big data giants like Baidu and Google, silicon vendors such as NVIDIA, Qualcomm and Intel's Mobileye, car OEMs like Mercedes, Porsche, and Nissan, to name a few.
Autonomous driving has clearly entered a very practical phase. The LiDAR ecosystem included companies that are specializing in cleaning nozzles and spray systems for the car sensors. There were vendors making transparent heaters that can be applied to keep the sensors free from frost and snow.
1. Weather and environment proof sensor placement. In today's cars, the windscreen can be kept clear in most weather conditions. These solutions can be reused for example by placing sensors in the windscreen area or by creating sensor cluster casings with familiar defrosting and washing solutions.
2. Keeping sensors clean and frost-free. This includes all new washing solutions, wipers, nozzles, spray systems, resistive heaters, and fans.
3. Monitoring and adapting to sensor conditions. It is necessary to monitor and diagnose whether the sensor is delivering high-quality measurements, and if it is not, the AI needs to support limited sensor driving modes using fewer measurements.
4. Vehicular communication systems. Vehicles have started communicating with other vehicles (V2V) and any objects around them (V2X). With this technology, cars will be able to announce their position and speed in real time and start "seeing" their environment independently of sensor measurements.
5. The new generation of real-time digital maps. Today, cars are getting new kind of map information: these maps are designed to be machine readable and include centimeter-accurate details of the roads, objects near the roads and other location data. Together with accurate outdoor and indoor positioning, these maps help cars navigate every corner and every small turn. Recently, our strategic partner around location-intelligence, HERE Technologies, shared insights about digital maps.
6. Enough data and the right data for training. Driving is complex. There simply is no AI solution that can generalize far beyond the data used for training. There must be enough (= lots of) representative data from all road conditions. Different AI models are needed depending on the weather conditions.
7. Taking the driver's seat is enough. As AI increases its capabilities, it is supposed to perform better than humans in more and more tasks. One day, driving will belong to one of these tasks: if humans can drive in winter conditions only by sensing the 360-degree view from the driver's seat, audible information, and acceleration of the seat, then AI should be able to do the same.
I'm confident that drivers and passengers can soon experience riding autonomous cars whilst enjoying movies, games or extra work time. However, it seems that one job remains for humans in winter conditions: brushing the snow off the car and scratching the ice off the windows before starting the engine.
Read more about Tieto's software engineering services for the Automotive industry.
Are you ready to share your car data?
Digital cars collect an incredible amount of data as they connect and interact with the environment. Many parties are eager to access and use your car data. Would you let them?
How will 5G and network slicing revolutionize business?
5G networks are being built extremely fast and all telco companies are competing on getting their networks ready to be piloted towards commercial use. One of the most interesting technical innovation is network slicing – the ability to slice the network in different "virtual networks" that are designed to meet the different requirements of usage, for example, real-time communications or services requiring high capacity.