2016 has seen the breakthrough of driverless car and truck technology into mainstream use. After many years of Google testing its autonomous cars, major car makers have started testing their own engineering prototypes and regulators have been addressing the legislative challenges. This has also led to start-ups being acquired and new, unexpected players entering the market, such as ride-sharing pioneer Uber.
The start of the year saw an A7 from Audi drive autonomously from San Francisco to Las Vegas, a 550-mile trip that kick-started the demonstration of driverless vehicles in the real world rather than the lab. January also saw Ford testing its autonomous engineering prototypes in Michigan in the snow, challenging the LiDAR laser and CMOS camera sensors to operate in highly reflective and dirty environments.
Volvo also started trials in Western Australia and announced plans to lease driverless cars to the public in Gothenberg, Sweden. The converted Volvo XC90 cars are using the DRIVE PX2 embedded processing card from NVIDA with the latest Parker processor to handle the fusion of the sensor data from cameras, LiDAR and radar. The Parker chip combines two of NVIDIA’s second generation 64bit Denver ARM-based CPU cores paired with four 64-bit ARM Cortex A57 CPUs. These all work together to provide up to 1.5 Tflops of performance alongside 256 of the latest graphics processor units (GPUs).
French start-up Navya rolled out its autonomous mass transit vehicle, carrying up to 16 people at a time around EDF’s nuclear power plant. The vehicle, also called Navya, uses two different types of LiDAR for detecting pedestrians and the road ahead. The shuttles run every three minutes, replacing several conventional busses and saving EDF over €3m a year in running costs.
Elsewhere in Europe two other mass transit autonomous systems are rolling out. The WEpod electric pods, designed by French manufacturer EasyMile for the Citymobil2 EU project, has already transported more 19,000 passengers in Vantaa, Finland, and Lausanne, Switzerland. And vehicles from Dutch system-maker 2getthere have also been on the road in the Dutch city of Masdar. These use virtual routes, defined in software, continuously calculating their position relative to their origin. The distance is measured by counting the number of wheel revolutions, and the position is calibrated using external reference points from simple, passive magnets embedded in the road surface. The small cylindrical magnets are spaced 2m apart and ensure the accuracy is within 2 cm on straight sections. This Free Ranging On Grid (FROG) navigation technology avoids the cost of physical guides such as rail or cables.
The pods already operate on public roads, but research continues to extend their use on all kinds roads and in different conditions for full autonomous operation. The €4m i-CAVE (integrated Cooperative Automated Vehicles) research program, led by the Technical University of Eindhoven, has been looking at how to link the 2getthere pods together to create a ‘virtual train’ with pods 0.3s apart using wireless links for a Cooperative Adaptive Cruise Control (CACC) system.
One of the biggest shakeups for driverless technology this year was the entry of Uber. It acquired the entire research group from Carnegie Mellon University and started to roll out driverless taxis in Pittsburgh. The vehicles, the Fusion from Ford and XC90 from Volvo, use the Carnegie Mellon software and while they still have a ‘driver’ who can take control in the event of an emergency, they operate autonomously. The Pittsburgh roll out follows Uber’s other entry into the autonomous vehicle market with Otto. The 91-person start-up develops systems and software for self-driving trucks, with staff from the self-driving development teams at Google, Apple, and Tesla.
The team is competing with Daimler-Benz, who have also demonstrated a self-driving truck. The future truck uses radar sensors linked to the throttle and braking systems to allow the trucks to follow each other as closely as a few metres, reducing drag from the air and boosting fuel efficiency. The front radar sensor has a range of 250 m and scans an 18-degree segment while a short-range sensor has a range of 70 m and scans a 130-degree segment. A stereo camera installed above the instrument panel has a range of 100 m, and it scans an area of 45 degrees horizontally and 27 degrees vertically. This monitors both single and two-lane roads, pedestrians, moving and stationary objects, information on traffic signs and even the road surface. The camera recognises everything that contrasts with the background, and so it can measure clearances of the top and sides of the truck precisely.
The US is not the only place where autonomous taxis are on the streets. NuTonomy in Singapore is rolling out driverless taxis using LiDAR, CMOS camera and radar sensors.
One of the things that had been holding back the testing and roll out of driverless cars has been the lack of legislation to support the technology. Up until now, a car or truck has required a driver. The Federal Automated Vehicles Policy from the Department of Transport in the US now allows for vehicles that can take full control of the driving task in at least some circumstances. Portions of the policy for highly automated vehicles (HAVs also apply to lower levels of automation, including some of the driver-assistance systems already being deployed by automakers today.
The guidance for manufacturers, developers and other organizations outlines a 15 point “Safety Assessment” for the safe design, development, testing and deployment of automated vehicles. The Guidance covers any organization testing, operating, and/or deploying automated vehicles, which includes traditional car makers and component suppliers as well as technology companies, start-ups or fleet operators who are customers of Autonomous Stuff.
The 15-point Safety Assessment outlines objectives on how to achieve a robust design. It allows for varied methodologies from Object and Event Detection and Response to Roadway Safety as well as Response and robustness of the HAV upon system failure. It also covers the validation methods for testing, validation, and verification of an HAV system, data recording and sharing requirements, post-crash behaviour and vehicle cybersecurity.
All the driverless cars at the announcement of these regulations in September used one particular technology supplier, called Autonomous Stuff. It’s ‘Automated Research Development Platform’ was used for the University of Michigan’s Mcity car and it supplied technology for the driverless cars from Carnegie Mellon University, MIT, Stanford, University of California Berkeley, University of Michigan and Virginia Tech Transportation Institute. It supplies sensors and middleware software such as Polysync, and this is being used by Kia for a self-driving Soul model.
Conclusion
2017 promises to be even more significant as the sensor and software technology matures. Apple has been developing technology for self-driving cars, and whether it will move into hardware or focus on software remains to be seen. Self-driving taxis and trucks will be rolling out across the world, with real world uses. That of course has led to problems. Google’s self-driving car has already had several crashes, and Tesla’s Autopilot, while not a fully autonomous control system, has also had problems with sensors leading to accidents.
However, a wide range of different autonomous platforms are mature enough in 2016 to be used commercially on public roads, is a huge shift. More will roll out in 2017, especially for mass transit, ready for autonomous cars and trucks to be available on the road in the 2020s.
