Top 10 Things A ‘Self-Driving’ Vehicle Must Do to Actually Be Self-Driving
Success in self-driving is not about confusing public perception or tricking consumers with hype. Despite what you might have heard, no vehicle for sale today is capable of SAE Level-4 automation—that is, a vehicle with autonomous driving capability, or the ability to drive without the need for human oversight or intervention operating under specific conditions. But, we are starting to see some availability of autonomous vehicles in both ride hailing and goods delivery services, some that are driverless and others still with safety operators due to the stage of development they are in.
To actually be self-driving, a vehicle needs to do more, much more than what the vehicles you’re able to buy or lease today are capable of. What follows is a list of crucial attributes of design and operation that are key to the delivery of a safe self-driving vehicle. These 10 attributes can help consumers better understand and evaluate where to place their trust. For a deeper dive into these areas, take a look at Argo’s newly released safety report. But if you just want the brass tacks, this list summarizes the basic requirements that every provider should meet for safe self-driving.
If a self-driving vehicle can demonstrate these capabilities, it is on its way to earning the trust it deserves. If it can’t, you shouldn’t trust it.
1) Sensor fusion for superior safety
Never mind what certain loud naysayers on social media have to say about lidar: using several different sensor types in partnership—namely lidar, radar, and camera—is the best way to ensure safety. The capabilities of each sensor complement the others, a concept that is particularly crucial when the driving environment gets complex. Dust, rain, and even entering or exiting a dark tunnel on a bright sunny day can temporarily disrupt the performance of a sensor. The Argo self-driving system (SDS) uses different types of sensors to enable the vehicle to continue driving safely and consistently even if the performance of one sensor is limited.
2) A solid backup plan
Engineers don’t design things to fail, but they do design backup solutions to work in case of failures—an engineering concept called “redundancy.” A well-designed self-driving system will use multiple versions of the key hardware components, whether that’s computers, sensors, or wiring. For example, Argo self-driving vehicles utilize two or more different types of sensors to oversee all areas around the vehicle so that if any one component fails, the others continue to function.
3) Dual, but different, computing systems
Equally important as redundancy is diversity in design, and to this end, the Argo SDS uses not just a main computing system, but also a backup system. The two systems work seamlessly together, while utilizing diverse software approaches, and different hardware components and processors, and serving fundamentally different purposes. By running each computer on independent and diverse software and hardware systems, there is a safe contingency in case one system fails.
4) Knows when and where it can play
A self-driving vehicle’s operational design domain, or ODD, is its offensive and defensive playbook. It defines the operating parameters of a self-driving system, and includes factors such as the speed limits, road types, and weather conditions in which the vehicle is designed to operate based on its sensor capabilities. To draw another sports analogy, the ODD marks out the field of play for the SDS, and defines the rules of engagement, just as the ball is only in play within the specific chalk lines of a ballfield. Argo uses the ODD to set the exact conditions and locations where its self-driving vehicles will operate safely to provide ride hail or goods delivery services.
5) Better maps = better driving
Self-driving systems use purpose-built, high-definition 3D maps, packed with intricate and highly specific street-level information to equip the SDS with everything it needs to enable good decision-making and drive like an experienced local driver. The importance of mapping and localization cannot be overstated; together, they give the SDS instant local knowledge—not just of the road layout and local highway regulations, but also where it is going. And the level of detail in a self-driving vehicle’s map affects how it will behave. Argo’s maps include locations of stop signs, crosswalks, traffic rule changes, and even which traffic lights control which lanes.
6) Practice good sensor hygiene
You know what can mess with a sensor? A lot of things. Water, mud, bugs, dirt, and bird droppings can degrade performance or even completely block a sensor’s view. The only way to overcome such sensor blockages is to consistently detect, diagnose, clean, and confirm the elimination of any obscurants. Working in close collaboration with our automaker partners, Argo designs its self-driving system to autonomously detect the need for sensor cleaning, and uses innovative devices such as air blowers, water sprayers, and even the redirection of heat for ice build-up to keep sensors unblocked.
7) Prepared for 3-headed monsters
The perception systems used by self-driving vehicles are trained to identify millions of labeled images, enabling them to recognize most of what the world has to offer—just consider the myriad different bicycles, scooters, motorcycles, cars, trucks, vans, and buses that we see on the roads every day. However, in the event the perception system doesn’t recognize an object, it still needs to be able to report the presence and motion of that unknown object. The perception system in the Argo SDS is designed to handle real-world situations, including the unknowns. If it sees a three-headed monster, as it might on Halloween, it may not know what it is, but it can still report that it sees an unknown object at a particular position and moving at a particular speed in a particular direction, and the vehicle can respond accordingly.
8) Follows both local rules and etiquette
Argo tests in multiple cities to ensure its SDS is exposed to a wide range of driving regulations, enabling it to operate appropriately and consistently with local rules, which often vary from place to place. Consider, for example, how a vehicle should behave when turning right if there is a bike lane. In California, a car may occupy the bike lane to turn right on red, but in Pennsylvania, the same right turn requires the car to stay in the vehicle lane. Argo’s powerful prediction system can incorporate a database of driving styles from which to match data, anticipate likely actions, make appropriate decisions, and avoid extreme situations in order to achieve “naturalistic driving.” The SDS can even handle the (in)famous “Pittsburgh left,” an unwritten rule in Argo’s home city which calls for oncoming traffic to give up the right-of-way and politely let left-turning vehicles turn against a green.
9) Detects and responds to emergency vehicles
It is vital that a self-driving vehicle knows how to safely behave in the presence of emergency vehicles and police responders. It should be able to detect and confirm sirens and flashing lights, and pull over safely until it is appropriate to resume its journey. The Argo SDS incorporates both microphones, to “hear” emergency vehicle sirens, and cameras powered by machine learning software, to “see” an emergency vehicle’s flashing lights.
10) Only goes when you’re ready
A well-designed self-driving system will prioritize safety above all else. For instance, the SDS will not depart from a ride-hailing scenario until all passengers are safely on board, doors are closed, and the “Ready to Go” button on the smartphone app or on the in-vehicle display is pressed. During the journey, we provide situational awareness and route progress information to the passenger(s). Our SDS will safely stop the car if someone opens the door while driving, or if someone hits the passenger-emergency stop button. Upon safe arrival at a drop-off location, the system will provide the passenger with a notification that the vehicle is in a safe state ready for them to exit.
In summary, Level 4 self-driving systems—those that are capable of operating without human oversight or involvement—are highly complex, integrated systems of software and hardware — which have to be far more capable than those delivering Levels 2 or even 3 of automation capability. The incremental nature behind the increasing levels of automation has led to a perception that belies the reality of achieving the difference between technology requiring human monitoring, and technology which doesn’t. While we’re starting to see Level 4-capable self-driving vehicles in ride-hail and delivery fleets, with more on the way soon, there are currently no such vehicles available for sale to consumers. What is available today is a wide variety of advanced driver assistance systems, or ADAS, which, when designed and used correctly, offer an enhanced level of convenience and safety, but don’t eliminate the need for a human driver.
The key to success in actual self-driving capability lies in the continuous and rigorous testing, monitoring, and expansion of a self-driving system’s capabilities to ensure that it is meeting the requirements of the road. Without the attributes detailed above, you can’t earn the trust of the public or meet the standards of safety that will serve the industry’s ultimate purpose: to improve road safety for everyone.