How to ensure the safety of Self-Driving Cars: Part 5/5

Congratulations for making it through the series! The last thing we’ll discuss is what it takes to make all this a reality. There’s lots of work still to be done to guarantee the safety of autonomous vehicles, so if you’re working in this field, good luck!

Part 5: What it takes to make self-driving cars objectively safe

In our opinion, there’s a few things that need to happen before we can objectively call autonomous vehicles safe:

Safety Criteria that we all agree on

One of the fundamental elements to this whole thing is agreeing on the metrics for vehicle safety. It must go beyond “x fatalities per 100M miles driven” into “The vehicle will make the correct decision 99.9999% of the time in emergency scenarios.” For all advanced technologies that have become ubiquitous in our world, there is some standards body that certifies them. The ISO and IEEE are the commonly used in the high-tech industry, and they have labs that verify that certain product meets their standards.

Figure 1: NHTSA Crash Test of Smart Fortwo (Source)

This doesn’t yet exist for the AV stack, and needs to exist to give the public confidence in the autonomous world. How are they going to verify the vehicle functions properly? Well, that could be anything from a set driving course with emergency scenarios, or some type of MIL/SIL/HIL simulation.

Just like a motorcycle helmet can’t be used on public roads without a “DOT” sticker, an autonomous vehicle shouldn’t be able to operate at level 4 or 5 autonomy (or potentially level 2 or 3) without a standards organization saying it’s safe.

More data

This one’s simple, we just need to log more miles. More miles on the road, more miles in simulation. One thing proposed is the sharing of data among vehicle makers. While this would greatly accelerate improvements in autonomous technology, it would cause some companies to lose their competitive advantage. Call us communists, but we’re of the opinion that the sharing of vehicle data among the industry is for the betterment of the world.

Standardization of the AV Stack

This one’s a toughie. Ultimately, there needs to be some standard software architecture that autonomous driving companies can leverage, similarly to how AUTOSAR (Automotive Open System Architecture) created the standardized software architecture for the automotive ECU (Electronic Control Unit) that controls vehicles today. This will ensure that the safety is guaranteed regardless of the autonomous application. If AUTOSAR isn’t the one to do this, I am sure another conglomerate of Autonomous Vehicle companies will bring this into fruition.

Figure 2: Technology Landscape for Autonomous Vehicles, Vision Systems Intelligence, LLC

Part of that standardized software stack will be the vehicle dynamics control system described in the planning section. This will be critical for allowing engineers to claim that they have created objectively safe vehicles that can perform better than humans in emergency scenarios.

Tools!

Here’s where we play. There must be simple to use tools that help Engineers deal with all this data and make their cars better. If you’re passionate about this too, shoot us an email and we should chat!

Conclusion

Thanks for your time. We would love to hear your feedback on this series. Let us know if we forgot something important, got something completely wrong, or even better, if you loved it. Feel free to connect with us on LinkedIn or via twitter or email.

For those working on making autonomous vehicles a reality, we salute you!

Read the Rest of the Series: How to ensure the safety of Self-Driving Cars

Part 1 — Introduction