Advanced Computing in the Age of AI | Thursday, March 28, 2024

BMW Backs Open-Source Platform for Driverless Cars 

<img style="float: left;" src="http://media2.hpcwire.com/dmr/autonomousBMW.jpg" alt="" width="95" height="64" border="0" />Since the first electronic driver assistance systems such as anti-lock brakes were installed in vehicles of the 60’s, we’ve come a long way. But in order to make this possibility a reality for millions of drivers, BMW says that current software platforms just won’t cut it. And instead of developing a proprietary platform, BMW is going all in on an open-source alternative, and isn’t afraid to explain why.

Since the first electronic driver assistance systems such as anti-lock brakes were installed in vehicles of the 60’s, we’ve come a long way. Already autonomous vehicles are being tested on highways around the world, and this week one consumer survey found that 90 percent of its respondents would buy an autonomous vehicle, provided their insurance rates were cut by 80 percent.

But in order to make this possibility a reality for millions of drivers, BMW says that current software platforms just won’t cut it. And instead of developing a proprietary platform, BMW is going all in on an open-source alternative, and isn’t afraid to explain why.

Although the auto industry currently sits on the step of developing a partially automated driving system where human drivers are still required to oversee the vehicle and step in if necessary, the long-term goal is to develop a vehicle that requires no human monitoring. But this means that the software driving the car must be highly complex to handle a myriad of possible traffic conditions, and highly reliable in order to keep passengers safe.

As BMW notes, this means pushing software beyond its current level such that it acts and reacts in a way that mimics an alert, attentive human brain.

To this end, BMW has outlined its autonomous vehicle platform requirements in a paper authored by Lukas Bulwahn, Tilmann Ochs and Daniel Wagner.

First, the authors note that although the software must be sophisticated, it should also also be approachable and uncluttered. Next on the list is computing performance, of which BMW is looking for a hardware-independent programming interface, symmetric multiprocessing, and performance-optimized processor architectures.

On the issue of safety, the researchers have pointed to software standardization, using software components with integrity levels up to ASIL B, as well as failure and misbehavior detection for hardware and software components.

But a large part of safety depends on lag times, or a lack thereof. Most drivers have at least one moment when split-second reactions save lives, so BMW has set the maximum reaction time variance to 100 microseconds, would require a timing violation detector, and would implement real-time scheduling to meet these timing requirements.

In addition, the automaker specified that systems must be designed with automotive standards in mind, and must be easily updated as computer vision and artificial intelligence technologies advance.

But none of these requirements inherently suit open-source platforms more so than a proprietary alternative, so what’s BMW’s reasoning for opening up to the public? According to the authors, the success of other open-source projects such as the Linux kernel and Apache HTTP are the first things we should look to.

Although none of these projects have had to conform to safety standards like autonomous cars will have to, they say that going open-source is still a better choice than re-implementing general-use commodity software for such a complex job.

There are already three “open-proof” projects in the works, which the authors established as a model for autonomous driving.

The first project involves creating open-source tools to verify that open-source software is being used correctly. The second, which involves developing a software kernel for the European train control system, is a development process conforms to industry safety standards, which will be essential for autonomous vehicles. Finally, the third addresses existing open-source software in hopes of certifying it based on safety standards.

From this, BMW has sketched out in their paper a plan for using Linux as the platform’s base, along with the AUTOSAR software component template for application software component reuse.

Although the authors will soon focus on expanding upon autonomous vehicle platform requirements, they say their ultimate goal is to unify the industry with this open-source platform “before others are engaged in closed-source, vendor-specific and limited platforms.”

EnterpriseAI