Baidu and automotive industry introduce principles for safe autonomous vehicles

Baidu and a coalition from the automotive industry have released a set of guiding principles for autonomous vehicles (AV), promoting a system of “safety by design” as conversations about self-driving cars go mainstream.

The coalition includes Daimler, BMW, Intel, Volkswagen, Fiat Chrysler Automobiles, Audi, and automotive supplier Continental, among others.

“In addition to offering broader access to mobility, [automated driving] can also help to reduce the number of driving-related accidents and crashes. When doing so, the safety of automated driving vehicles is one of the most important factors,” the group said, explaining the motivation behind the principles.

AV proponents have pointed out that the vehicles could end up being safer than human-driven cars. However, the coalition highlights a number of topics that need to be resolved in order to meet this goal.

The nearly 150-page document covers 12 areas including cybersecurity protection, driver-vehicle handovers, data recording, coping with component failures, and awareness of an autonomous system’s limitations. According to the group, balancing safety and availability in Level 3 and Level 4 autonomous vehicles, those that require human interventions in certain scenarios, is difficult to balance.

AV safety will peak when operations are optimized but restricted to certain driving scenarios, the group says. However, if restrictions are too high, safety goals cannot be reached due to limited availability. The same is true when limitations are too liberal.

“Being too risk-averse leads to a system that is overly conservative, and the system availability becomes too low, which in turn will not provide the benefits of a safer and more comfortable customer experience,” the report said.

The principles also highlight the importance of cybersecurity. A slew of possible safety risks arise from malicious actors seeking to take advantage of the connected vehicles, which could possibly allow them to gain access to a car’s controls.

In June, cybersecurity firm Regulus Cyber was able to spoof a Tesla’s GPS system to redirect it off a highway. However, the researchers had to put an antenna on the vehicle in order to launch the attack.

Similarly, in April, Tencent’s Keen Security Lab was able to trick a Tesla into switching lanes. The researches put stickers on the road to fool the vehicle into altering its behavior. Though the attack didn’t require any hacking, it highlights how AVs could be manipulated if safety issues are not thoroughly assessed and resolved.

The principles draw attention to a reliance on data, whether gathered by sensors or provided by maps and GPS, in order for AVs to function properly. “If the integrity or authenticity of this data is compromised, the building blocks of the automated driving functions will use faulty data to maneuver the vehicle, which might result in inaccurate driving or other deviations from correct operation,” it said.