Knight Rider Meets Uber

Dear Impossible Readers,

As much as I enjoy fast cars, drifting, and the thrill, I value efficiency and productivity more. So, I take the train to work. Why? Well, because it is less stressful. I can eat my breakfast if I had a bad morning, read a book, or write some emails, and start my workday while commuting. More importantly, I can do all of that without being stuck in traffic. We almost always need to queue for every type of service you can think of, so why would I queue in my car behind other vehicles? What if we no longer needed to do that?

Autonomous cars do not require magic, but they must be aware of their surroundings. Vehicle-to-vehicle and vehicle-to-everything technologies enable cars not only to react but also to coordinate actions. Communication between vehicles allows them to share information about their intentions, such as when they plan to brake or merge, or when they detect hazards like icy roads ahead. Awareness and communication do not need to be mutually exclusive. They perform well individually, but together they form a powerful combination, making travel smoother, safer, and smarter.

Currently, autonomous driving relies mainly on sensors. Cameras, lidar, radar, and ultrasonic detectors enable vehicles to perceive their surroundings and respond accordingly. Tesla’s Autopilot, Waymo’s robotaxi services, and Mercedes’ Drive Pilot all use a combination of cameras, radar, and lidar to identify other vehicles, pedestrians, and obstacles. This sensory awareness is already powerful enough to enable cars to operate autonomously, but a driver still needs to remain attentive. The latest systems utilise advanced deep-learning architectures to combine data streams efficiently. Bird’s-Eye-View fusion projects sensor inputs into a top-down view, facilitating scene understanding by merging lidar’s depth accuracy with camera detail. Transformer-based models, like TransFusion, assign dynamic weights to different sensors depending on conditions, adaptively adjusting importance when one sensor is less effective. Techniques such as Dual Perspective Fusion Transformers integrate raw radar and camera data to perform better in adverse weather such as fog or rain, while thermal imaging enhances night-time reliability. These systems depend on precise localisation and mapping (SLAM) processors that align data with GPS and high-definition maps, capable of handling millions of data points in real time through specialised processors, balancing accuracy and latency. In essence, they combine diverse sensors, machine learning, and high-performance computing to enable autonomous vehicles to perceive, predict, and plan safely in complex environments.

Challenges in autonomous driving include processing speed. Deep learning models for perception and prediction are computationally intensive but require response times within milliseconds. To address this, engineers prune and quantise models and deploy specialised accelerators such as NVIDIA’s Drive platforms or Tesla’s FSD chips to reduce latency without compromising accuracy. Additionally, sensors must be reliable in various conditions. Cameras fail at night, lidar struggles in fog or rain, and radar can generate clutter in dense environments. To mitigate this, vehicles employ redundant sensor fusion, probabilistic filters like Kalman or particle filters, and adaptive fusion models such as TransFusion or Bird’s-Eye-View approaches, which adjust reliance depending on sensor reliability at a given time. Furthermore, communication barriers need to be overcome. For vehicles to coordinate effectively, they need to “speak the same language,” yet DSRC and C-V2X remain competing technologies. Industry groups like the 5G Automotive Association are advocating for C-V2X as the standard, with city-scale pilots already testing cooperative manoeuvres and intersection management.

Ultimately, this shift is not just about cars driving themselves. It is about building a network where every trip becomes smarter, safer, and more sustainable. This could enable platforms to share autonomous vehicles. Such a platform would require the vehicles to adapt and optimise their routes flexibly. AI can combine traffic data, passenger requests, and vehicle capacity to create efficient routes with fewer empty seats and shorter waits. In areas with good public transport, vans could supplement existing systems by managing highway or feeder routes. In regions with poorer networks, they could fill gaps and reduce reliance on private cars. Beyond convenience, fewer vehicles on the road means lower emissions and a more sustainable transport system.

Ready, steady, go,
Yours Possibly