What California Regulatory Filings Reveal about Robotaxi Operations at Scale

Imagine a city where the cars that pick you up have no drivers. They are not just electric vehicles; they are fully autonomous, capable of navigating complex streets, stopping for traffic, and finding their way to your destination without a human hand on the wheel. This is the promise of the robotaxi, a technology that has moved from science fiction to the streets of cities like San Francisco and Los Angeles. The hope has long been that these machines would solve several problems at once. By being electric, they would cut pollution. By being shared, they would reduce the number of cars clogging the roads. And by being automated, they would operate with a level of efficiency that human drivers cannot match, perhaps by keeping cars moving constantly and never needing to park while waiting for a passenger.

However, the reality of how these fleets actually operate on a massive scale has remained a mystery, hidden behind company dashboards and private data. For years, the public has relied on summaries, simulations, or models to guess how these systems work. But a new analysis has peeled back the curtain, using official government records to see exactly what is happening. Researchers at the Lawrence Berkeley National Laboratory examined thousands of pages of reports filed by robotaxi companies with the California Public Utilities Commission. These filings, required by law, contain detailed logs of every trip, every mile driven, and every time a vehicle stops to charge. By sifting through this data, the researchers uncovered a picture of robotaxi operations that is far more complex, and less efficient, than the optimistic vision of the future.

The study focused primarily on Waymo, the largest operator of driverless taxis in the United States, and compared its performance with Zoox, another major player currently running a large-scale pilot program. The researchers looked at the data from January 2025 through June 2026, a period when Waymo had grown to a fleet of roughly 3,285 vehicles in California, accounting for about 80 percent of the entire national fleet. They tracked how many miles these cars drove with passengers inside versus how many miles they drove empty. In the world of ride-hailing, driving empty is called "deadheading." It is a necessary part of the business, as cars must drive to pick up a passenger and then drive away after dropping them off. But the goal of automation was supposed to be minimizing this waste.

The data tells a different story. While Waymo did improve its efficiency over time, the empty driving remained stubbornly high. Between early 2025 and mid-2026, the share of miles driven without a passenger dropped from 53 percent to about 42 percent. While this is an improvement, it is still significantly higher than what is seen in human-driven ride-hailing services, exceeding those figures by more than 50 percent. The researchers found that even as the fleet grew and the cars got better at finding passengers, a massive amount of energy was still being spent on moving empty vehicles. This persistent empty travel runs counter to the idea that automation would automatically lead to a leaner, more efficient system.

Perhaps the most surprising finding concerns what the cars do when they are not carrying a passenger. The researchers discovered that a large portion of the "empty" time is not spent driving around looking for a ride, but rather sitting still. As the fleet expanded, the total amount of time vehicles spent waiting—parked or idling—grew by 39 percent. This suggests that the system is not as flexible as a human-driven network. When demand spikes, human ride-hailing services can draw on a vast pool of drivers who are already on the road or can quickly log in to work. Robotaxis, however, rely on a fixed fleet of vehicles that must be physically present to serve a request. To handle a sudden rush of passengers, the system needs a large number of cars sitting in parking lots or depots, ready to go. This creates a massive "parking footprint," where thousands of vehicles are stationary for long periods, consuming space and resources without moving.

The energy implications of this behavior are significant. The researchers used the frequency of charging sessions reported in the filings to estimate how much electricity the fleet was actually using. They found that the energy required to keep the fleet running was far higher than what the cars' official efficiency ratings suggested. For Waymo, the energy use implied by the charging data was roughly double the vehicle's rated propulsion, reaching about 69 percent of the energy a typical gasoline-powered car would use for the same distance. This extra energy is consumed by the constant need to move empty cars, the time spent idling, and the frequent charging cycles required to keep the fleet operational. When they looked at Zoox, the energy overhead was even higher, reaching about 147 percent of the gasoline-equivalent benchmark.

The study also highlighted a shift in how the empty miles are being driven. In the early days of the service, most empty driving was "unassigned," meaning the cars were cruising around without a specific destination, hoping to find a passenger. Over time, this changed. A larger share of the empty miles became "approach" miles, which are the trips taken to get to a passenger after a ride has already been requested. This suggests that while the cars are getting better at finding passengers, the distance they have to travel to reach them is increasing, possibly because the service area has grown or because the cars are parked further away from where people are asking for rides.

The researchers were careful to note that these findings are based on real-world operational data, not simulations. They examined millions of trip records and charging logs, providing a rare, concrete look at how these fleets function at scale. The data does not support the idea that robotaxis have already achieved the efficiency gains many expected. Instead, it reveals a system that is still struggling with the fundamental challenges of moving empty vehicles and managing a fixed fleet in a dynamic environment. The promise of a cleaner, more efficient future with robotaxis remains possible, but the path to getting there is not as straightforward as simply turning the steering wheel over to a computer. The data suggests that without significant changes to how these fleets are managed—perhaps by better matching cars to demand or by finding ways to reduce the need for massive amounts of stationary vehicles—the energy and space costs of robotaxis could remain high.

Ultimately, this analysis serves as a reality check for the robotaxi industry. It shows that while the technology is advancing and the fleets are growing, the operational realities are complex. The efficiency gains from automation are being offset by the need to keep a large number of vehicles ready and waiting, leading to substantial energy use and empty driving. As these services continue to expand, the focus may need to shift from simply adding more cars to figuring out how to make the existing fleet work smarter. Until then, the vision of a perfectly efficient, driverless future remains just that—a vision, rather than a reality.