Opening AV Tech Doors with Open-Source Software

Open source software solutions are already being deployed globally by automakers investing in autonomous vehicle technologies.

The potential for open source software applications in everything from simulation and testing to programs to operation and infotainment is fostering a sense of collaboration among the industry’s top performers, a trend many say is only likely to accelerate. “If you look at the autonomous vehicle industry, all the leadership people running the platforms, many of which came from academia who are looking at this through the lens of shared research and open source, which allows you to do more with limited time and budgets,” explained IDC analyst Matt Arcaro.

He noted open source has always a component of building an open and modern IT infrastructure and use of open code for autonomous vehicles today is driven by the need for the industry to advance quickly and develop systems that are safe and secure. “There is a definitely a role it plays but, if you look at having a completely open source autonomous vehicles that would run a solution end-to-end, that is too tall an order in the near term,” he said. “At some point, there would be enough trust for that but an end-to-end view of how they’re built, maintained and updated, that is still a question for open source.”

Sohini Roy, product manager for the automotive supplier at Canonical, which produces Ubuntu is a Linux distribution based on Debian, argues the use of open source software provides autonomous vehicle developers with three advantages: transparency, quality and velocity. “Ultimately, what it comes down to for those in the autonomous vehicle space, we’ve seen time is really money and making computers that can make nuanced decisions is a huge challenge,” she said. “When we think about what open source is, it’s freedom to study the program, to distribute, to learn and to reuse, that allowed those three elements to come together and maximize price performance.”

Arcaro says one of the biggest roles open source could play is in employing artificial intelligence and machine learning model development, ranging from simulations to data visualization, within data sets and training sets. A growing number of car companies have made their autonomous vehicle datasets public in recent years, a trend kicked off by Daimler in 2016 when it made its Cityscapes dataset freely available.

That move was followed in 2018 by Baidu sharing its ApolloScapes datasets, while Aptiv published its nuScenes datasets that same year while Lyft, Waymo and Argo followed suit in 2019. “Open source has gained a lot of traction and the vast majority of solutions you’re developing have a lot of commonality,” Arcaro explained. “Through open source development in general, you see a broad-based set of collaborators that you can build quickly on top of.”

He noted it also helps with groupthink, more differentiated feature sets and leveraging capabilities from outside the organization. “It helps companies start to focus on areas where they see gaps in privatized solutions,” Arcaro said, “Open source starts to allow companies to look at the situation in a new way, solving what’s unsolved and building the frameworks and capabilities around it to meet requirements.”

Even automaking giants like Ford and Audi have begun releasing datasets from their AV research programs to the public, adding to the growing sense that a shared ecosystem for AV development can be, and should be, founded on open source. “We’ve essentially become holistic partners with automotive business where open platforms can improve things from testing to automation to operating services to infotainment,” Roy said. “Open source has been critical in pushing the boundaries of innovation in all those fields.” Roy pointed to automaker investment in open source on the simulation environment development side, noting the deployment and sharing of data sets, and making them open, allows you to get advanced quality assurance as well as access to more testing scenarios, including the more challenging ones.

Arcaro pointed to use of platforms like Tensorflow, an end-to-end open source machine learning platform that is underpinning a lot of the models being used for AV perception and actuation planning. He also highlighted openpilot, an open source, semi-automated driving system developed by comma.ai, where the development is supported by an open-source community using Discord and GitHub.

Arcaro noted among the biggest hurdles to broader use of open source software architectures is security, followed by functional safety. “You need to understand vulnerabilities and the ways to update those when they are found, which requires monitoring and operations software investments to make sure there aren’t exploits,” he said. “On the foundational level, there’s challenges around aligned timelines and goals building something together. The more parties you involve, the more likely it will take longer and your voice could be minimized.”

Roy said she hopes Ubuntu Core, which is built for embedded environments and focused on security and reliable updates, will address concerns like data encryption, and the availability of the systems. “A combination of features like OTA updates and tamper proof systems allows us to push critical updates as fast as possible in a way that automakers want, in a way that is properly QAed and examined,” she said. Other approaches like isolation, which prevent processes from interfering with one another, minimized privileges and mixed criticality are all ways to allow open source drive innovation in less safety critical environments, Roy noted.

Arcaro predicts an increased focus on open source will be the trend in 2021 and beyond as AV developers large and small try to find ways to advance open source software, as opposed to developing their own systems. He sees this already happening in China with the Baidu Apollo open source platform, which has already been used by a considerable number of players for their hardware and software stacks.

“The auto industry will continue see the value of open source and you’re already starting to see them embrace it more than they have,” Arcaro said. “There is value there and there is some comradery from the immensity of the challenge and meeting deadlines, which requires working together in unique ways.”