Autonomous Driving: What We Need to Know

September 2016

Recently, I had the opportunity to engage with a few startups developing autonomous driving systems. This prompted me to explore the autonomous driving space, mapping out its critical elements and key players. My observations are drawn from discussions with startups, VCs, and software developers, and reflect my interpretation. Apologies in advance if I’ve overlooked any finer points.

Autonomous driving, at its core, is about data. It involves systems that gather data from various sources, process it, perform redundancy checks, and send feedback to control units to make decisions. From this perspective, autonomous systems can be divided into three functional areas:

1. Data Acquisition

This includes positioning, vision, and sensing technologies. Established tools like LIDARs, stereo cameras, and other sensors dominate this space. The focus now lies on reducing costs, improving form factors, and enhancing performance efficiency. Companies operating here are often Tier-1 suppliers, module makers, or startups specializing in sensor technologies.

2. Data Processing

Data processing represents the most challenging and opportunity-rich domain for autonomous systems. It encompasses several critical components:

• Operating Systems
  Companies like Google, Tesla, and Uber are pioneering distinct approaches. Google leverages LIDAR sensors combined with its robust mapping data to sense the environment, supplementing this with 3D cameras for redundancy. However, their reliance on cloud-streamed mapping data or massive onboard storage presents challenges. Tesla, in contrast, uses 3D cameras integrated with human input for a simpler, more intuitive solution, though it faces limitations with camera blind spots.

Startups such as Nutonomy, Comma.ai, and Advasworks are also actively developing operating systems. These companies might attract acquisitions by OEMs (Original Equipment Manufacturers) lacking expertise in AI, machine learning, and image processing.

• Mapping Engines
  Accurate mapping is vital for autonomous driving. While standard GPS-enabled mapping achieves a five-meter accuracy, autonomous systems demand precision down to 2–3 centimeters to avoid blind spots. This requires integrating sensor data with highly localized mapping in real time.

Google has a notable edge here due to its deep expertise in mapping. Recognizing this, German automakers BMW, Daimler, and Audi acquired Nokia’s digital maps business for $3.1 billion. Similarly, Uber bolstered its capabilities by acquiring Bing Maps and Decarta. Other players like Baidu, TomTom, Garmin, and Apple are also well-positioned to collaborate with OEMs or develop their own systems.

• Processing Units
  The massive data generated by sensors, cameras, radars, and LIDAR must be processed rapidly. Companies like NVIDIA, with expertise in building GPUs, have developed processors capable of creating real-time 3D models of a car’s surroundings. AMD and Intel are also poised to play critical roles in this domain.

3. Vehicle Control Technologies

This segment is more mature, with significant advancements already in place. Major Tier-1 suppliers like Bosch, Valeo, and Delphi dominate this space.

Emerging Technologies: Routing and Optimization

Routing technology is another critical piece of the puzzle. Autonomous cars will rely on advanced algorithms to determine optimal routes, a challenge akin to solving the “traveling salesman problem” on a city-wide scale. Companies like Uber and Lyft already leverage such technologies, and Google’s $1 billion acquisition of Waze highlights the importance of this expertise. As autonomous cars gain traction, routing efficiency will become a defining advantage.

Key Challenges and the Road Ahead

The primary bottlenecks in autonomous driving lie within the data processing ecosystem: interpreting terabytes of sensor data, ensuring real-time processing, and transmitting this data with minimal latency over robust networks (e.g., 4G/5G). Companies with expertise in AI, cloud computing, and sensor technology—such as Google, Uber, and NVIDIA—are better positioned to tackle these challenges than traditional automakers like BMW, Ford, or Nissan.

This raises an intriguing question: Will traditional automotive OEMs become the Pegatron or Foxconn of the future—relegated to manufacturing hardware for tech-driven systems? The answer remains uncertain. However, the evolution of autonomous driving will undoubtedly reshape the automotive landscape in ways we are only beginning to understand.