Becoming Swarm: How Our Approach to Self-Driving Must Change

For 20 years we have been promised that autonomy will create safer, more efficient roads, but the latest data shows we are headed in the opposite direction. Recent reports from the U.S. Department of Transportation (DOT) indicate intelligent driving systems may be causing as many crashes as they prevent.?

Big tech’s strategy involves dependence on their own servers and cloud data. Communication with those servers works wonderfully for ordering sushi, but the resulting one to two seconds of delay equates to many meters on the road – a level of error that can easily mean life or death. In order to measure safety and understand relative motion, we need timing to be accurate to tiny fractions of a second and we need positioning to be accurate to centimeters rather than meters.?

Many leaders in the space emphasize the role of onboard individual AI, which can help. Cruise and Waymo have done very well in achieving their goal of individual perception and their vehicles can drive for long periods without someone controlling the wheel. However, despite this impressive magic show, we have not yet been able to reduce congestion, increase saftey or limit green-house gas emissions. Many studies have shown that if your car must wait to respond to the car directly in front, it’s impossible to dampen the accordion effect responsible for the majority of congestion and the associated pollution. Using big-tech's combination of individual AI and centralized servers, U.S. traffic deaths have hit a 20-year high. Neither the centralized control from offboard servers nor the individual AI can accomplish synchronized motion which is essential for safe roadways.?

So what is the alternative? We must become swarm. We need to emphasize relative position more than global position and focus more on our neighbors than cloud-based maps. We need ground truth and accurate relative positioning to measure spacing and flow. Swarm intelligence does not seize control away from you or the AI you may have on board, but it does require good positioning, reliable connectivity, and the ability to measure safety and flow. To achieve that, we need positioning to be accurate to fractions of a meter and we need timing to be measured in fractions of a second. On its own, GPS cannot accomplish either of these feats reliably. Does it really make sense to reference satellites thousands of kilometers away to understand cars that may be only a meter apart?

Most people don’t understand how vulnerable we are from over-dependence on global positioning and cloud-based connections. If GPS gets us to the Dunkin' Donuts we are content. The problem is that the error of GPS in cities can easily be 30 meters, which prevents us from achieving the goal of safer, more efficient roads. GPS jamming is increasing especially in Ukraine, Russia, Norway, Finland and Taiwan. The National Oceanographic and Atmospheric Administration (NOAA) warns that military action is not the only danger, explaining that natural solar activity can knock out GPS for long periods.?

Recently, the CIO of the Pentagon warned that we must look beyond GPS. The best option today may be the use of distributed, peer to peer networks based on IOT devices such as the emerging Ultra Wideband and Long Range LoRa radios already entering millions of phones and cars. This is a realization of our previous work in the military to enable navigation in so called GPS-denied situations. Ash Carter, secretary of defense under Obama, called for a “GPS of things.” The swarm positioning he envisioned could democratize positioning and communication, taking back our individual data rights and ensuring more reliable, accurate data flows through our systems.?

Today's centralized network of cell towers, the cloud, and the high definition maps are controlled by tech giants who parade self-driving as the next big thing that justifies their ability to control and monetize our individual data –the greatest source of revenue in history.? Bill Gates loves going for rides in self-driving cars and says they’ll come first as taxis and commercial trucks sometime within the next decade. What he doesn’t explain is that the motivation is less about shortening your commute, and more about inserting a sales associate into your backseat. Meanwhile, the tech giants have burned through billions developing cars that seem to drive by themselves, but actually depend on off-board satellites and servers. If we were not stuck in this quagmire, perhaps safe, efficient roads would be making headlines like ChatGPT has been.?

In 1997 I wrote a remedial chatbot and using NSF grants I created a fully autonomous vacuuming robot. This early robot brain went on to be used across many defense robots. At the time, I thought conversational AI was the stuff of science fiction and much harder than navigating a car or robot. Two years later, working with the Defense Advanced Research Projects Agency (DARPA), we focused mostly autonomous driving, thinking it was the low-hanging fruit. Now it seems NLP technology has won the race to large-scale societal impact.?

Desired reductions envisioned in the twenty-year-old DOT Vision Zero plan have not been achieved and may never be if we don’t change tack. I’ve been making autonomous vehicles for over twenty years and I see swift and graceful roadways in the future. However, the future we want cannot emerge unless we shift to emphasize the human-robot team within the car. With the human front and center we can shift from "self-driving" to "collaborative driving" where initiative from you, your car and your neighbors is coordinated based on relative motion and measurable safety.

The alternative is that we fail to measure safety and allow marketing hype to rule the day. Tesla claimed they were in the lead, but they have now recalled their self-driving software. In 2021, Tesla accounted for 70% of the smart cars involved in crashes and, perhaps worst of all, their system was shown to stop autopilot assistance in the last moments before a crash --a design decision that seems more about avoiding litigation than saving lives. The blame lies not on one tech company, but rather that we placed our faith in the wrong place. We confused cause and effect. If we have accurate relative positioning, autonomy becomes a much easier problem and measurable safety helps us adapt and improve. The inverse is not true. Increasing individual autonomy does not necessarily produce better flow or safety.?

Forget about self-driving cars for a moment and consider what swarm thinking could mean. A significant study by the state of Michigan provides insight into the power of swarm thinking. Initially, they wanted to prove that slowing down would improve safety and congestion. They lowered the highway speed limit to 55mph and stepped-up enforcement. When they measured the impact of their hypothesis they were shocked to find that accidents and congestion had worsened. It turns out that humans are complicated. About half of drivers dropped their speed from the old speed limit of 65mph to the new limit of 55mph. The other half matched the speed of those around them, generally maintaining the 75mph they felt was safe. These two groups conflicted as the increasing disparity in speed equated to increased congestion and accidents.??

If we want harmonious roadways we need to focus on flow, and this means that synchronicity is the goal… not rule following. AI drivers may be the worst rule followers of all time. Neither centralized control nor individual intelligence can accomplish flow. The traditional way of thinking would assert that everyone should just have followed the 55mph speed limit, but the well-intentioned state troopers had already done the best they could by stepping up enforcement. Perhaps the most impressive thing about this study is that the Michigan leadership adapted. Within a short time after analyzing the data the speed limit shot up to 75mph and since then flow has improved. From a practical perspective this achievement began not with autonomy but with accountability and objective data. The first step to improving our roadways should be measurable safety – objective motion data so we can know what is really happening.? We can’t improve what we can’t measure.

About the author

David Bruemmer is founder of W8less which is a leading provider of AI and robotics. David has led pioneering robotics and self-driving programs for the Army, Navy, the DOE, DARPA and the Department of Transportation. David received an R&D100 award for developing the first AI system fielded on military robots. The latest version is now a reusable, portable AI brain used on many kinds of vehicles. Other firsts include working on a 100 robot swarm for the DoD, a drone that can land autonomously on a moving car and an autonomous robot that can keep up with fast moving soldiers for an entire day.