Making Sense of AI - Part 5

?

Self-Driving cars and unSYSTEMatic growth

I’ve been learning about self-driving cars recently, an excellent example of cutting-edge technologies that represent a fascinating integration of computer vision, sensor fusion, control systems, and machine learning. They hold potential benefits in reducing human error in driving, improving traffic flow, and increasing accessibility for those unable to drive. But as I learn more, I’m left pondering some big questions and curious about how these vehicles will interact with and perhaps transform the complex systems of traffic management and navigations, especially in the Indian context.

First things first, what are Self-driving cars?

Self-driving cars, also known as autonomous vehicles, utilize a combination of technologies like LiDAR (Light Detection and Ranging), radar, cameras, and GPS to navigate and respond to their surroundings.

LiDAR sends out laser beams and measures how long it takes for the light to bounce back after hitting objects. This helps the car create a very detailed map of its surroundings, so it knows where things are, like other cars, people, trees, and more.

Radar uses radio waves to find out if there are objects nearby and how quickly they're moving towards or away from the car. This is really useful for a self-driving car to understand things like whether the car in front of it is moving fast or slow, or if it's stopping.

Computer vision uses cameras to see the world, just like our eyes do. But it's not just about seeing; it's also about understanding what it sees. The car's computers analyze the images from the cameras to recognize different things like road signs, traffic lights, pedestrians, and other vehicles.

Complexities and challenges of developing self-driving car systems

Reliable Interpretation and Prediction in Varying Conditions:

Self-driving cars must operate in diverse environments, from quiet suburban streets to bustling city centers, and in all weather conditions. This requires AI systems capable of accurately interpreting sensory data (like images, LiDAR, and radar signals) to recognize and react to a vast array of objects, road conditions, and human behaviors.

Predicting the actions of other road users (cars, cyclists, pedestrians) is a major challenge, anybody who’s driven on Indian roads will testify to that! It involves not just recognizing their current state but anticipating future actions. This requires advanced modeling of human behavior, traffic flow, and situational awareness.

Training AI Models on Extensive Datasets for Edge Cases:

Machine learning models used in autonomous vehicles are trained on massive datasets comprising real-world driving scenarios. However, the rarity of 'edge cases' – unusual, unexpected events or conditions – makes it difficult to ensure that the AI has been exposed to enough of these scenarios during training.

Let’s consider this in the context of Indian roads where we see an amalgamation of cars, trucks, cycles, scooters, pedestrians, rickshaws, autos, tempos, tractors, traffic lights that don’t work, traffic policemen who don’t show up when you need them, cows, street beggars, more cows, potholes, cars parked on the main road, jumping the red light, using the mobile phone while driving, driving from the wrong side and still confidently giving death glares to others, jumping the lanes or not acknowledging the existence of lanes, honking like there’s no tomorrow, motorcycles bypassing lane traffic by getting on the sidewalk, Uber drivers who simply don’t care, did I mention cows? I don’t see us having edge cases, I think we ARE the edge cases.

Perhaps it won’t be so bad to have self-driving cars that follow the rules. But since technology is always adopted at a gradual rate, I can’t help but imagine a poor little self-driving car sitting in the middle of traffic chaos, not knowing what to do.

Redundancy and Fault Tolerance:

Critical systems in self-driving cars, like those responsible for navigation, decision-making, and control, must have redundancy. This means that if one system fails, another can take over, ensuring continuous safe operation. Fault tolerance is not just about hardware but also software. The system must be designed to gracefully handle failures or unexpected inputs without causing unsafe conditions.

As an aside, I’m worried about navigating the social dynamics in a world where we only have self-driven cars. Some of us take cars and driving as a religion. If we don’t drive, who are we really? What happens to our YOLO road trips? Do we lose the joy of rolling our eyes and poking fun at the women drivers? I think we are all united by one thing, and that’s a universal belief that each one of us is better than the next driver on the road. What happens when two self-driving cars collide, what is the expectation from the passengers sitting in the back? Do we get out and have a good old-fashioned fight, or do we wait while the machines ‘gracefully handle failures’? Will we need new hand gestures to show our frustration? (asking the real questions here??)

Cybersecurity Concerns:

As connected vehicles, self-driving cars are susceptible to cybersecurity threats. This could range from data breaches to more severe risks like unauthorized control of the vehicle. Ensuring robust cybersecurity measures is crucial to protect against hacking, maintain user privacy, and ensure the integrity of vehicular systems. I certainly do not wish my daily ride to become a Black Mirror episode.

Regulatory Challenges:

Current traffic laws and regulations are based on human drivers. When self-driving cars become a reality, the legal framework will need adaptation to accommodate autonomous vehicles, addressing liability in accidents, insurance models (can’t wait for this one!), and the setting of safety standards. I wonder what the driving license of the future will look like. Whose photo will it show?

Impact on Urban Planning, Environmental Policy, and Transportation Economics:

The widespread adoption of self-driving cars will significantly impact urban infrastructure, in terms of road design, traffic management, and parking structures. I’m especially interested in parking structures, it seems very enticing that cars will park themselves because parking gives me more anxiety than a balloon in a cactus garden.

Not so sure about the environmental impact. Will they contribute to lesser emissions, especially if integrated with electric vehicle technology and optimized routing? There's the potential for increased vehicle kilometers traveled, which could negate these benefits.

Economically, the shift could disrupt current transportation models, affecting everything from public transit systems to the taxi industry. It also opens opportunities for new business models like on-demand transportation services. Which again brings up the question, do we really need more cars on the road? Are we trying to fix a broken system by creating a new one? It’s important to think about whether we are treating symptoms, or the disease.

With so many challenges, is there any hope?

In the end, we are only as good as the perceptions we hold. In order to build trust, we definitely need more awareness, transparent communication about the capabilities and limitations of AI in vehicles, public demonstrations of safety features, and clear information about accident response mechanisms, not to mention the overall environmental impact and impact on employment. At any rate, I can get behind the more humane aspect of increasing accessibility for those who can’t drive themselves for whatever reason. And automatic parking ??

It’s exciting technology but one that requires a thoughtful blend of progress with pragmatism. ?Let us not shy away from the gritty details. As we stand at this crossroads, the path we choose must be guided not just by technological capabilities but by a vision of sustainable, inclusive, and practical mobility solutions.?