Waymo vs Tesla: Two Strategies for Conquering Autonomous Driving

Waymo and Tesla, the two main forces in the race for autonomous cars, have chosen radically opposite strategies. Srikanth Thirumalai, vice president of software at Waymo, described this in an interview with Bloomberg.

Waymo's full-stack control

Waymo builds autonomy from the bottom up. The company develops its own software and its own algorithms, and controls the hardware itself. It is a full stack with no external dependencies, neither on chip makers nor on third-party perception modules.

The key tool here is its own simulator. Instead of driving a car on real roads every time, Waymo tests scenarios in a virtual environment. It is faster, safer and cheaper. The company can model rare, dangerous situations that happen on real roads once in a million kilometers.

But the main piece is the safety-critique system. This is not just a code check. It is a system that critically evaluates the decisions made by the autonomy agent. Before the car carries out an action, the system checks whether it is safe. This adds a layer of protection between perception and execution.

• In-house development of software and algorithms
• Simulation instead of repeated real-world testing
• A safety-critique system as a second line of control

Tesla: reality as the teacher

Tesla took a different route. It does not try to create perfect software in a lab. Instead, the company released FSD (Full Self-Driving) into the real world, into the hands of hundreds of thousands of drivers. Every trip is data. Every braking event and every turn is a training signal.

Tesla builds on cameras. No lidars, no radars, only vision, like a human. This is simpler and cheaper at scale, and data flows in constantly from the entire fleet at once.

The FSD beta let Tesla collect millions of hours of real-world driving. It is a huge dataset that Tesla uses to train neural networks. Every bug and every edge case becomes an improvement for all the other cars. A refusal to open the doors in a snowstorm? The next update will account for it.

Two philosophies, two speeds

Waymo is slower but cleaner. It will check a thousand times before releasing to the masses. Tesla is faster but takes risks. It learns from mistakes in real time, and not all of those mistakes are harmless.

Waymo has the advantage in safety and control. Tesla has it in scale, iteration speed and data volume. These are two logics for solving the same problem: how to teach a car to drive without a driver.

What this means

Autonomous cars will arrive not thanks to one approach, but thanks to both. Waymo will show the industry how to do it safely and reliably. Tesla will show how to do it at scale and profitably. The winner is whoever first combines both: the safety of the lab with the data of real roads.