Unmanned Systems Technology 026 I Tecdron TC800-FF I Propellers I USVs I AUVSI 2019 part 1 I Robby Moto UAVE I Singular Aircraft FlyOx I Teledyne SeaRaptor I Simulation & Testing I Ocean Business 2019 report

10 Platform one June/July 2019 | Unmanned Systems Technology Researchers at Stanford University have developed a way of controlling autonomous cars that integrates prior driving experiences to boost safety (writes Nick Flaherty). Its machine learning algorithm is based around road-tyre friction data. It has been tested on a racetrack using Niki, an autonomous Volkswagen GTI, and Shelley, an autonomous Audi TTS. The system performed about as well as an existing autonomous control system and an experienced racecar driver. “We want autonomous vehicles to work in many scenarios, from normal driving on high-friction asphalt to fast, low-friction driving in ice and snow,” said Nathan Spielberg, a researcher in mechanical engineering at Stanford. “We want our algorithms to be as good as the best skilled drivers – hopefully better.” Instead of just relying on real-time evaluation of the environment, the new algorithm includes data from recent manoeuvres and past driving experiences, including trips Niki took around an icy test track near the Arctic Circle. The road-tyre friction data dictates the limits of how hard the car can brake, accelerate and steer in order to stay on the road in emergency scenarios. If engineers want to safely push an autonomous car to its limits, such as having it plan an emergency manoeuvre in icy conditions, this data is essential; unfortunately it is highly variable and often difficult to predict. The neural network for the controller integrates data from past driving experiences at Thunderhill Raceway in Willows, California, and a winter test facility that provided 200,000 physics- based trajectories. “With the techniques available these days, you often have to choose between data-driven methods and approaches grounded in fundamental physics,” said Christian Gerdes, professor of mechanical engineering at Stanford. “We think the path forward is to blend these approaches in order to harness their individual strengths. Physics can provide insight into structuring and validating neural network models, which in turn can leverage massive amounts of data.” In comparison tests, Shelley first used a physics-based autonomous system, preloaded with information about the course and conditions. When compared on the same course during 10 consecutive trials, Shelley and a skilled amateur driver generated comparable lap times. The researchers then loaded Niki with the neural network system. The car performed similarly, running both the learned and physics-based systems, even though the neural network lacked explicit information about road friction. In simulated tests, the neural network outperformed the physics-based system in both high-friction and low- friction scenarios. It did particularly well in scenarios that mixed these two conditions. “With so many self-driving cars on the roads and in development, there is a lot of data being generated from all kinds of driving scenarios,” said Spielberg. “We wanted to build a neural network because there should be some way to make use of that data. “If we can develop vehicles that have seen thousands of times more interactions than we have, we can hopefully make them safer.” Prior experience required Driverless cars The Shelley autonomous Audi TTS being tested at Thunderhill Raceway Park in the US