Fast-Tracking Safe Autonomous Vehicles with Simulation

Eric Bantegnie

If the experts are right, 10 years from now our automobiles and trucks will drive themselves, and delivery drones will fill the sky to hasten the arrival of the latest book we are dying to read or the pizza we are craving to eat. Robots will be doing some of our jobs, so we might have more time for both of these activities.

With so much autonomous machinery in our lives, we need to ensure that they all safely operate together, so drone-delivered pizzas are not raining on our heads and autonomous cars are not running into robots—or people. The only way to do that in this complex environment of the near future is through engineering simulation.

Consider the case of autonomous cars: RAND Corp. estimates that they will require approximately 8 billion to 11 billion hours of road testing to be certified safe. One major competitor in the race to market reports that they test drive autonomous vehicles about 0.4 million miles per year. At that rate, engineers will need four to five centuries to do sufficient testing to guarantee our safety. Obviously, that’s not realistic if automakers are promising delivery in the next five to 10 years.

Using simulated driving scenarios, the same company is performing virtual road tests at a rate of 8 million miles per day. That’s remarkable, but it is by no means a limit—simulation is scalable. We can greatly increase the number of simulated road miles per day by adding more compute cores to tackle the problem and exploring more driving scenarios. Scalability makes the challenge of 8 billion to 11 billion miles of road testing manageable.

For drones, safely navigating the three-dimensional open spaces of the skies will require that they avoid their fellow drones on their frantic delivery runs, as well as power lines, buildings and other obstacles as they approach the ground. Robots—depending on their application—may have to climb stairs, run into a burning building to save humans or manipulate a delicate machine part without damage.

Extensive Sensing Capabilities Required

All three of these cases—autonomous cars, drones and robots—require remarkable sensing and actuating abilities to function efficiently and safely. Radar, lidar, infrared radiation and visual wavelength cameras must be designed so that autonomous devices can sense their environment at high enough resolution to do their jobs without fail. Photorealistic virtual reality that simulates light and human vision will add to the sensing capabilities and safety of autonomous vehicles. Signals from one device must not interfere with another, whether there are two autonomous cars driving on the same road in a small town or hundreds of them on a superhighway during rush hour in a major metropolitan city.

Simulation can help from the component level all the way up to the complex system of cars during rush hour. Engineers can use electromagnetic simulation to determine the shape and range of a radar signal coming from a single unit sitting on a laboratory bench, and also when that radar unit is installed behind the fascia of a car’s bumper with tons of steel, plastic and composites behind it. Simulation can map the possible interference of these signals with each other in a traffic jam; it can also help cars to communicate with each other so they can know what the other is planning to do—unlike human drivers. The number of lives lost to traffic accidents will plummet when autonomous systems are reacting logically to incoming data without emotion or human error.

That logic will come from the software embedded in the hardware of these autonomous devices. And, that too, can be simulated, to great effect. Although the operating system that controls your computer may seem fragile, crashing at the least opportune moment, it doesn’t have to be that way. Importantly, for safe autonomous systems it cannot be that way: They must work the first time and every time.

Software generated by simulation can be tested for flawless operation to meet the most stringent certifications in the world, including ISO 26262 for automobiles, DO-178C for drones and Asimov’s Three Laws of Robotics (just kidding on the last one). Simulation can ensure that the software that controls the hardware that is out of our control will be safe and reliable for all involved parties.

Whether we are dealing with autonomous driving, flying or maneuvering, simulation will play an enormous role in bringing safe autonomous vehicles to market—in five years, not 500.

Eric Bantegnie is ANSYS (ANSYS.com) vice president and general manager. Send comments on this commentary to [email protected].