AV Technology - Facts vs. Fiction
Waymo Remains in the Pole Position, While Tesla is a Show Me Story.
The surge in expectations surrounding both public and private autonomous vehicle (AV) technology companies echoes the electric vehicle (EV) SPAC and IPO frenzy of 2020-2021, which peaked with Rivian's late 2021 IPO. At the time, bullish sentiment swept across the EV sector, fueling exaggerated expectations about the speed and scale at which new EV companies would emerge. Investors and analysts projected rapid market penetration and profitability, overlooking the immense difficulty of building an auto company from scratch and achieving the production scale necessary to compete in an industry defined by capital intensity and operational complexity.
The fallout from the EV hype cycle has been severe. Fast forward to today, most of the EV SPACs that went public during that period have either gone bankrupt or are in a distressed financial state. In the U.S., aside from Tesla, Rivian appears to be the only EV OEM with a potential path to survival — and that has come at the cost of burning through billions of dollars.
A similar pattern is now unfolding in the AV space. Analysts, investors, and the media are overemphasizing corporate press releases with limited, out-of-context performance data, often overstating the significance of new driver-assist systems and misrepresenting them as being close to fully autonomous solutions. This has fueled overly optimistic projections that numerous companies are on the brink of successfully developing fully autonomous vehicle systems.
These projections underestimate the technological challenges involved in building a driverless vehicle system. Just as scaling an EV company proved far more difficult than initial projections suggested, it’s likely that only a handful of companies in the U.S. and Europe will succeed in commercializing Level 4 systems — and eventually Level 5 systems capable of operating in all driving environments and conditions.
The Technical Challenges of Autonomous Driving
In the foreseeable future, the U.S. robotaxi market is likely to consolidate around a small group of dominant technology enablers—likely three or four key players—capable of developing fully autonomous, driverless systems that can reliably operate across diverse driving environments and conditions. Waymo and Tesla are leading this race, albeit with different approaches to autonomy.
Substantial technological and capital barriers to entry in autonomous vehicle development have already forced companies like Cruise, Motional, and Argo AI to scale back operations or exit the market. This underscores the formidable challenges of establishing a commercially sustainable autonomous vehicle system and robotaxi service.
Driving is an open-domain problem with an infinite long tail of complexity and unknown scenarios. Autonomous vehicle systems must handle countless edge cases — unpredictable, low-frequency events, from changing weather conditions to unexpected obstacles to the unpredictable behavior of other drivers.
The core challenge for autonomous vehicle systems is not just recognizing objects within the driving environment but understanding the environment itself. Object detection and pattern recognition alone is not enough — systems need to infer the intent and behavior of other road users and adjust their responses accordingly. This requires more than mathematical function approximation; it demands elements of general intelligence, such as abstract reasoning, real-time adaptation, and interpreting the physical environment in a nuanced and context-aware manner to make optimal driving decisions. Bridging this gap still requires advances in decision-making algorithms, diverse data sets, and sufficient onboard sensor fusion and compute resources to support real-time inference.
A recent drive in a Tesla Model Y highlighted a key technical challenge for autonomous vehicle systems. The drive was through a familiar suburban environment, and overall, the FSD system performed well — except for two disengagements.
One disengagement occurred when the vehicle struggled to navigate out of a parking lot. The second disengagement, however, was more noteworthy and informative. While driving through a suburban village, the vehicle came to a complete stop in the middle of an intersection despite a green traffic light. The vehicle’s halt seemed to be triggered by FSD detecting a pedestrian standing at the corner, waiting to cross the street. However, the pedestrian was stationary, the traffic light was green, and the pedestrian signal showed a red "Don't Walk." Despite these cues, the vehicle stopped unnecessarily. This is a good example of an autonomous system identifying an object in its environment but failing to understand the broader context needed to make an optimal driving decision.
Tesla’s vision-only system correctly identified the pedestrian but lacked the contextual understanding to conclude that it was safe to proceed. A more robust sensor suite could have improved performance in this situation. Radar, for example, measures an object’s distance, speed, and relative motion — key data points that would have confirmed the pedestrian was stationary and not about to cross, prompting the vehicle to drive through the intersection safely.
This example illustrates a current limitation of Tesla’s vision-only approach. In contrast, Waymo’s lower disengagement rate reflects the advantages of its more comprehensive sensor suite — combining cameras, radar, and LIDAR to provide richer and more diverse environmental data. Tesla’s reliance on a vision-only approach could prove to be superior in the long run, but in the near term, it still struggles to match the decision-making consistency of a multi-sensor system.
Perception vs. Reality of Autonomous Driving Technology
The gap between advanced driver-assistance systems (ADAS) and fully autonomous technology is significant, a point often overlooked. While Level 2+/3 ADAS systems offer "eyes-on, hands-off" convenience, they fundamentally differ from a fully autonomous system. This misconception is commonly seen in the misrepresentation of new ADAS products, such as BYD’s “God’s Eye” system, as being technologically close to fully autonomous solutions. Such conflation leads to an inflated perception of how close many companies are to deploying driverless systems, while simultaneously underestimating the formidable technical hurdles that remain.
This misperception contributes to an exaggerated sense of progress, downplaying the inherent complexity of achieving true autonomy. The technological gap between Level 3 and Level 4 systems is often understated, while the transition from Level 4 to Level 5 represents an even more significant technological hurdle.
While Chinese companies are advancing in autonomous technology, some of this progress is also overstated. Many operational robotaxi networks in China are confined to fixed route services, revealing these systems limitations compared to the broad capabilities required for navigating diverse, unstructured driving environments.
Context Matters in Measuring AV Progress
Outside of China, there is a significant opportunity for third-party suppliers of autonomous vehicle systems. One such promising player is Wayve, which is positioning itself as a hardware-agnostic autonomous vehicle system supplier. Wayve, like Tesla, is pursuing an end-to-end driving system.
However, assessing autonomous technology companies, both private and public, remains challenging due to limited performance data and detailed information. This challenge is compounded by promotional company announcements that often lack the context needed to evaluate system performance or benchmark it against competitors.
For example, Wayve recently reported approaching UK performance levels in the U.S. after collecting just 500 hours of incremental U.S.-specific training data. While this sounds impressive, the significance is hard to gauge without baseline intervention rates and more quantifiable data for context.
This situation echoes the EV SPAC bubble, where investors and the media overemphasized corporate press releases, neglecting thorough analysis of a company's technology strategy and commercial viability. While Wayve may ultimately succeed, such announcements perpetuate the misconception that numerous AV players are on the verge of success, overlooking the substantial technological and operational hurdles that remain.
Mobileye’s autonomous vehicle system development strategy contrasts with Wayve’s and is more aligned with Waymo’s “compound AI system” approach — combining engineered elements with extensive use of neural networks. A compound AI system reflects a willingness to make trade-offs to achieve reduced generalization error, utilizing defined operational parameters to lower disengagement rates.
While AI technology may eventually advance to a point where a pure end-to-end neural network system demonstrates superior performance, current data supports the notion that a compound AI system still delivers better real-world performance and a lower rate of critical interventions.
Waymo Remains in the Pole Position, While Tesla is a Show Me Story
Waymo maintains its position as the autonomous technology leader, underscored by its first-mover advantage and pragmatic "compound AI system” approach that enables a more robust and commercially viable robotaxi service. This strategy, combining high-definition mapping, rigorous safety protocols, and modular AI frameworks with extensive use of neural networks, ensures greater reliability—a critical factor for commercial deployment. Waymo's approach not only reflects the most viable technology today but also provides a flexible foundation for future advancements including the eventual adoption of an end-to-end AI system as the technology matures.
Tesla's management team continues to pursue a more complex technological path, relying heavily on an end-to-end neural network system. This approach, as discussed in my post The Ultimate AI “Killer App”, still faces significant technical challenges in handling real-world variability and edge case scenarios, which could take considerable time to overcome.
The recent steep correction in Tesla’s stock reflects fundamental weakness in EV sales. As highlighted in my post What’s Next for Tesla, although Elon Musk’s political activities may have damaged the brand and impacted sales, the deeper issue is Tesla's stale and narrow product lineup — a fundamental headwind that predates Musk's heightened political visibility.
There are emerging expectations that the anticipated June launch of Tesla’s robotaxi service in Austin could be delayed by a couple of months. This wouldn't surprise close Tesla observers. While EV sales remain weak, the market narrative is likely to shift back toward autonomy, and could trigger a rally in its share price ahead of Tesla’s robotaxi launch. However, any more significant delays in launching Telsa’s robotaxi service or performance disappointments would likely trigger another wave of selling pressure.
Tesla's reliance on an end-to-end neural network system, and the still existing technical challenges inherent in this approach, could impede and delay a successful driverless robotaxi launch. This concern is reinforced by data from the Tesla Full Self Driving (FSD) Tracker — though an imperfect dataset — which still shows frequent FSD disengagements, indicating that the system is not ready for an unsupervised, driverless deployment.
Tesla may continue to present its system as an end-to-end neural network, however to mitigate these risks, Tesla may need to adjust its near term strategy to more closely resemble Waymo's approach. This would necessitate adopting a "compound AI system" that integrates more engineered elements with neural networks and focus the initial rollout and expansion to additional cities to fixed geographical areas. Tesla already plans to launch in a defined area in Austin, which makes sense, as unsupervised FSD is likely to perform better on familiar, frequently traveled roads.
While Tesla may continue to pursue its long-term goal of a pure end-to-end neural network system, this pragmatic shift could accelerate commercialization, improve reliability, and increase the chances of a successful deployment. Without this strategic adjustment, Tesla risks ceding further ground to market leader Waymo.