There is much talk about Artificial Intelligence these days, but what we don’t often think about is how pervasive A.I. has been ever since the first computers were added to our vehicles. It’s all about defeating inertia and closed-loop control.
Overcoming inertia
The opening premise of what I call static inertia is that any object not in motion will remain at rest unless energy – either external or internal to that object – applies physical force to overcome the static state. Whenever we open the door to any vehicle, we’ve applied human intelligence along with muscular and skeletal force to overcome inertia, as we open the door and slide into the driver’s seat. Our next humanly intelligent action is to turn the key or apply the brake and press the START button. At that point, electrical energy from the vehicle’s battery is released through switching devices to begin a process of defeating inertia. We give little thought to this unless it doesn’t happen to work the way we expect it to.
On most vehicles (even hybrids), the engine’s static inertial state is overcome by a large amount of electrical current applied to an electric motor as it begins to spin the internal combustion engine (ICE). Then, atomized fuel is delivered to each cylinder. An ignition events begins the combustion process and this drives the pistons, connected by rods to a crankshaft. If the engine is mechanically healthy and responds to fuel and ignition, the engine runs on its own and the electric motor has done its work.
Of course, as soon as we turn on the ignition key or press the button, the onboard computers on modern vehicles instantaneously wake up and begin to process inputs. This is in preparation for providing the engine with fuel and ignition in the proper time and quantities as soon as it begins to spin. This is “snapshot” fast. Thus, engine control units can be considered artificially intelligent because they receive pertinent inputs, perform calculations, and apply outputs with very specific goals using closed-loop systems for optimal efficiency.
Older computers were clunky and slow with their processing, and had only one or two closed loops, but newer computers do their processing much faster as today’s vehicle dynamics demand it. But even with their clunky slowness, some older engine control computers tended to employ closed-loop reasoning and apply remedial A.I. logic by (for example) enriching the fuel mixture and disengaging the A/C compressor if they determine that the engine is beginning to overheat.
When the starter begins to spin, the static inertia of both the starter motor and, by extension, the engine has been overcome. The dynamic inertia of work-producing motion has been created, fueled by atmospheric nitrogen selectively superheated above the pistons. This is carried out by carefully timed combustion events. The point here is that Artificial Intelligence makes it work right.
Then, through shafts and complex gearing, the transmission or transaxle receives engine-turning torque. It then multiplies it and delivers the torque to the wheels via a final drive system. Thanks to gravity, the drive wheels are resting on the ground, and thanks to friction, when the driven wheels begin to turn, old-fashioned inertia that keeps the vehicle sitting still is overcome and the vehicle begins to move in whatever direction the wheels are driving it. Steering is largely done with human intelligence, but A.I. can take over steering in some cases, as in automatic parking.
On any moving vehicle, dynamic inertia must be dealt with by the brakes to stop the car. But applying the brakes too rapidly can cause a loss of steering control, so anti-lock brake systems use a closed-loop A.I. system (remedial though it may be) to isolate sliding wheels, pulsing the brake fluid pressure at those wheels, effectively enabling the driver to maintain control of the vehicle.
More advanced vehicle dynamics systems will have sensors telling the computer when the vehicle is in an oversteer or understeer situation and will apply brakes automatically to selected wheels to alleviate it. This is possible because of additional inputs and algorithms, moving up a bit on the A.I. scale. Also, the PCM may briefly de-tune the engine at the point of transmission shifts and use “fuzzy logic” to prevent unwanted transmission downshifting while speed control is used.
If a driver accelerates too rapidly and the drive wheels begin to spin, most of the newer platforms will use ABS braking to automatically apply one brake, stopping the the spin, and even command the PCM to limit electronic throttle response to prevent wheel spin. It can even use that same countermeasure to prevent excessive highway speed beyond the OEM tire rating. In other words, these strategies and others like them represent algorithms the onboard computers employ together to protect drivers from their own poor choices. That’s a form of A.I.
Driver intelligence and reaction
As intelligent as an advanced driver assist system (ADAS) might be, with its cameras, radar, lidar, etc., the intelligence of the driver makes decisions for the vehicle from the moment he or she slides into the driver’s seat. We as human beings, using our external inputs, engage in our own closed-loop processing without thinking of it as such. We use our eyes and ears for input information, and we control the vehicle using the steering wheel and our feet on the pedals as outputs.
When we hear and feel the engine start, we know it’s time to apply the brake or the clutch pedal and put the vehicle in gear if we’re ready to drive away. If the cabin is too warm or too cold, we work the controls on the dash panel to bring the temperature to a comfortable level. If we intended to go backward and we feel ourselves going forward, we stop the vehicle and make corrections. The speedometer is the feedback we observe that tells us if we’re driving according to the prevailing speed limit. Of course, we can choose to ignore the speed limit, but some GPS systems will issue us a warning about it.
Another feedback loop we humanly use is our perception of what’s happening in the outside world (i.e., what we observe through the glass and in the rearview mirror). We watch the brake lights and turn signals on other vehicles to determine what those other vehicles are planning to do. But if we can’t see an approaching vehicle or if another vehicle does something we don’t expect, the vehicle we’re driving may contact other vehicles with varying levels of damage. If we’re following another vehicle too closely and that vehicle has to stop very suddenly, the brake lights on the vehicle may not give us sufficient time to react and there may be a crash. Center high mount stop lamps (CHMSL) help, but only so much, even though an operational CHMSL reduces rear-end collisions by 40 percent.
Then there are those times when we’re driving on a road we travel literally every day and we’re so comfortable that we’ve settled into the complacency of believing we know what to expect on that road until we encounter a vehicle stopped where no vehicle has ever stopped before. In those moments, we can find ourselves in the middle of a panic stop and/or a collision. Rounding a familiar curve or topping a hill on a country road to find a vehicle just sitting there is a harrowing experience.
Sometimes during city driving, we may be thinking the driver in front of us is going to “squeeze the orange light” before it changes to red. We can try to follow suit at that moment, and again, we might suddenly find ourselves in a panic stop situation if that driver suddenly changes his or her mind and decides to stop at the light instead of going on through it. Many of us have experienced that.
If we’re not paying attention to the outside world (texting) or if we can't see well because of heavy rain, or fog or maybe because of the sun shining across a pitted or dirty windshield, we may pull out in front of other vehicles, run into objects in the road, crash into objects beside the road, or drift into the wrong lane. I’ve seen white vehicles driving in heavy rain without any lights on that were dangerously invisible and also driving in my blind spot. A house trailer being pulled out of a side road on a foggy morning can almost guarantee a nasty crash.
Animals and people sometimes do unpredictable things. A deer, a dog, a horse, or a hog may unexpectedly cross in front of us. I’ve personally seen deer run across a crowded road in a busy town at night – I almost hit one. And we’ve all seen jaywalking pedestrians step from behind a parked vehicle just as we’ve seen kids scare the daylights out of us by chasing a runaway soccer ball.
People may cross the road in plain sight while fiddling with their phones or while in absent-minded thought, like one construction worker I saw who finished a conversation with a fellow hard hat, then turned and walked right out in front of the vehicle I was test driving without even thinking about the fact that he was crossing a busy highway. Fortunately, I was able to predict his movement and stop because I was watching him very closely. That’s why we as intelligent drivers need to obey traffic laws and pay close attention to the road and everything near our path of travel at all times.
Driving too fast is not intelligent, but it’s something we can decide to do because we’re in a hurry or craving the excitement of high speed. But even driving the speed limit on slippery roads can be dangerous. And parking lot maneuvers need to be executed very cautiously because other cars and pedestrians are everywhere in those situations. Newer cars have parking aid alarms and onboard video to help with this. Backing out of a downtown parking place while watching for oncoming traffic can cause us not to notice the vehicle directly behind us waiting in line at the next traffic light, and parking aid monitor/video feed can help with that.
Conventional wisdom and statistics both dictate that errors in human judgment are the main factor of accidents. ADAS systems use low-level A.I. to help.
ADAS and Derq® A.I.
ADAS addresses a set of user services categories that are driving manufacturers’ research and development:
Safety — directly contributing to vehicle safety:
• rear-end collision warning
• roadway departure warning
• lane change/merge collision warning
• intersection collision warning
• railroad crossing collision warning
• vision enhancement
• location-specific warnings
• collision notification
Safety Impacting — potential to distract or aid the driver:
• navigation and routing
• real-time traffic information
• driver comfort and convenience features
The Intelligent Vehicle Initiative (IVI) is a multiagency research and development project intended to accelerate research, development, and use of OEM or aftermarket in-vehicle systems to help drivers operate more safely and effectively. Of course, V2X Communication is a huge part of this effort. Eventually, all new vehicles will be able to communicate with other vehicles, infrastructure, the overarching network (through cell phone towers or directly through the internet), and even with pedestrians’ cell phones so that a pedestrian can be warned by his or her phone if they are in danger from an oncoming vehicle (This is what is meant by V2X). Liability-prone carriers like UPS, FedEx, etc. can add onboard communication units such as the Kapsch CBX-9360 for vehicle-to-infrastructure communication.
I interviewed Jaime Sullivan who is Derq VP of Sales, and Jaime who explained how Derq traffic analysis software is collecting metadata (using existing infrastructure works) to provide data not previously available on such a large scale. And the best information comes from cameras at intersections. Cell phones can’t provide reliable real-time information due to latency. One of the problems is that the local infrastructure must be integrated with the vehicle(s) in question for the system to work at peak efficiency. Traffic cameras can be used by the vehicle-to-infrastructure network by way of software written for that purpose. Derq is creating software that will one day be embedded in the infrastructure and new vehicles so each vehicle can communicate to drivers threats in their immediate vicinities. One example would be vehicles that are approaching a traffic signal too quickly to stop. Other examples would be pedestrians, bicyclists, and even large animals that are crossing the road. Older vehicles can be retrofitted with equipment as the customer desires, but it won’t be required.
Early warning in these situations is what Derq is working on. Their stated goal is detecting, tracking, and classifying road users. Pilot programs are being tested in several states, i.e., Nevada, California, Michigan, Ohio, Washington, and Colorado, as well as in other countries. Again, Derq software identifies targets, their projected trajectories and speeds, and how those targets will intersect as they travel. And existing assets are presently used for the best results; the best data available isn’t pulled from connected vehicles at this point because intravehicular data can’t predict what the vehicle sensing units can’t see (like a vehicle coming from around a corner). If, for example, a vehicle is traveling at a speed that indicates it will pass through a red light, the onboard vehicle systems can’t help much, but Derq software can interface with the infrastructure to enable the traffic light, signaling perpendicular (endangered) traffic to remain red a bit longer, preventing a possible accident.
The Derq system provides cities with metadata concerning near-miss situations that happen frequently so the city managers can analyze traffic and craft solutions. This isn’t a new concept. Every month for about 10 years I traveled through Valdosta, Georgia, on Highway 84, and in those years, if you set your vehicle speed to the city speed limit, you'd hit every green light through the main part of town. I’ve driven in a lot of cities and that’s the only place I’ve ever seen that. The point is that it didn’t happen by accident, and somebody did some serious work to make it happen.
So where is the software loaded? Well, Derq is modular and very flexible so it can be deployed in a Department of Transportation data center, on an edge device at an intersection, or run in the cloud and be used on live, historical, or semi-real time data from sensor feeds in the city. Then using Derq software and analytics to sort out the gathered data either in real-time or later. The endgame is that there will be a version of Derq or some competitor running in every vehicle and most cities to enable perception and intent prediction from not only the vehicle but also from outside of the vehicle systems for all road users.
