Autonomous Mobility in Dense Urban Environments
The Inevitable and Unavoidable Path To Level 5
If you’ve ever driven a car for long periods of time you know how much easier it is to drive on a freeway versus city streets. The limitless and unforeseen chaos lurking on a city street, especially in highly populated, very dense cities like New York or Los Angeles gives you a glimpse into the level of detail required to ensure a self-driving car reacts properly in any given situation at any given moment. Freeways are exactly that, free of distractions that a driver would typically experience and need to pay close attention to on the streets of New York City.
Just think for a moment of all the movement simultaneously occurring on any typical main street in New York City. Pedestrians, skateboarders, j-walkers, people on scooters, people on bikes, people heading in the wrong direction, pedestrians getting out of cars, pedestrians getting into cars, trucks, taxis, the potential for a pedestrian to jettison out in front of a random car at any given moment and so many other countless scenarios. The unpredictability becomes vast and almost infinite. Preparedness is not only significant but critical and essential. Reaction time is of paramount importance. Inevitably, this is the training ground for level 5 autonomous mobility.
The scenario as depicted by the photo above lends itself to machine learning, where the technology used to “train” the automobile to “drive” like a human through the use of highly sophisticated camera technology (coupled with other sensors for like radar and lidar) that captures the data in real-time and then allows the developers to then feed the data back into the vehicle enabling it with the capacity and capability to learn and grow each and every time becomes a great advantage, and I would argue a premier method by which advancements become exponentially easier while simultaneously delivering unprecedented learning growth.
Why Machine Learning/Deep Learning?
Why machine learning/deep learning? Machine learning/deep learning essentially automates “prediction” through perceiving the environment through human eyes via camera sensors, similar to how a human driver has to make smaller, yet safety-critical predictions while driving at all times. It’s the ability to predict the behavior of the objects in the immediate surroundings while driving in the surrounding, something humans do without giving a second thought to the action.
What’s more, the technique or methodology of machine learning/deep learning enables the computer/controller to learn how a human would learn, through sensation and feedback which creates a continuous loop of data inputs and outputs, consistently improving with enhancements at every stage (if things are progressing correctly) and every scenario based on the last set of inputs or the last data points from the previous driving environment meaning it allows for enhancements by “…automatically applying accurate and responsive correction to a control function.” The entire process requires a multifaceted toolkit to enable the type of continuous progression made on a consistent basis. The proper CPU power, the proper data center for data storage and integration, along with artificial intelligence.
Infrastructure Is of Paramount Significance
Infrastructure becomes a necessity and is the “glue” the binds everything together. Without the proper infrastructure in place that is capable of handling the eventual communication between automobiles, traffic lights, signals, and a host of other devices in what will be known as “smart cities”, then the “autonomous” in autonomous mobility will likely cease to exist. Invariably, then, most cities will begin to revamp entire transportation infrastructures in order to usher in a new mode of transportation that will include not only self-driving cars but autonomous trucks and the like. The future is paved in technology overhauls that will require everything from new traffic lights, to street sensors and detection systems that simultaneously and seamlessly communicate and share data with one another.
The process of progressing from level 3 advanced driver assistance systems (ADAS) to full autonomous mobility (level 5), one that requires the removal of input from a “driver” (meaning no brake pedal, no accelerator pedal, and no steering wheel) mandates an unprecedented amount of planning and overhaul of city infrastructures at a national scale. Inevitably, all of the requirements may seem daunting, but let’s not forget all of it is a minimum of thirty (30) years in the making from this point forward.
My name is Patrick Salem — I am an autonomous mobility professional, engineering and project manager for self-driving cars. I’ve worked in automotive autonomous mobility platform development and strategy, including platform design, systems architecture design, requirements development, commercial aircraft electronics systems development and scope of work documentation. See my other articles on Medium at AutoDriveAI and follow me on Twitter, Instagram, and LinkedIn.
Medium: AutoDriveAI
Twitter: AutoDriveAI
Instagram: AutoDriveAI
