In a facility located in Utah, a large group of Tesla employees is diligently reviewing video footage captured by vehicles employing the company's Full Self-Driving (FSD) technology. The footage includes instances of the cars colliding with various animals such as cats, dogs, and deer, as well as more typical driving incidents. There are concerning reports of vehicles failing to brake before impact and often exceeding speed limits. Workers have also observed near-collisions involving children playing nearby.
These individuals, referred to as "data labelers," play an essential role in enhancing Tesla’s AI-driven driver-assistance systems. They meticulously annotate both positive and negative driving behaviors and escalate any significant issues to engineering teams tasked with improving the technology.
Elon Musk, the CEO of Tesla, claims that FSD will soon achieve full autonomy for all Tesla vehicles. However, discussions with nine former data labelers and a past engineer focused on self-driving reveal that recent iterations of the software continue to struggle with fundamental driving tasks, such as yielding to emergency vehicles and stopping for school buses when picking up or dropping off students.
Despite these critical limitations, Musk and other Tesla executives have underscored FSD's safety in their communications, pushing for public demonstrations of the fully autonomous capabilities that Musk has consistently assured investors would manifest for a decade. One notable demonstration was the launch of a robotaxi pilot in Austin, Texas, last June, which included some human monitors in the cars and others monitoring remotely.
Before the Austin rollout, the company spent considerable time mapping the area to prepare for the robotaxi service, contradicting Musk’s claims that the software would function effectively in real time anywhere. Four former Tesla employees disclosed that in the lead-up to these demonstrations, staff worked extensive hours to map routes and train the software to navigate specific hazards, making the self-driving technology appear more competent than it might genuinely be. These preparatory efforts cast doubt on Musk's assertion that Tesla’s technology could scale globally without extensive localized mapping, something that rivals employ.
In terms of safety claims, Tesla often reiterates that its FSD system is significantly safer than human drivers, with some executives positing that it is up to ten times safer. However, an investigation by Reuters challenges the integrity of these statistics, revealing gaps in how Tesla compares its data to federal accident statistics. For example, the company appears to inflate its safety metrics by only comparing FSD-enabled vehicle crashes with serious incidents that include airbag deployments, skewing the data relative to broader, less severe accident statistics.
Tesla has faced a series of federal investigations and lawsuits linked to accidents involving its self-driving technology, including several high-profile fatalities, prompting scrutiny from regulatory bodies like the National Highway Traffic Safety Administration (NHTSA). The agency is currently overseeing multiple investigations into FSD’s readiness, including incidents wherein the system failed to respond correctly to traffic signals and emergency situations.
Amidst these criticisms, former Tesla employees voice significant concerns about the FSD system's safety and reliability. Many express a lack of trust in the technology, with some stating they would not feel comfortable being passengers in a robotaxi.
Tesla's data labelers meticulously inspect FSD's performance, watching footage of the system making critical errors, such as failing to yield to emergency vehicles and crashing into concrete walls after neglecting to brake. Some reported instances of FSD-equipped vehicles approaching construction zones too hastily and nearly colliding with workers on site.
Access to FSD failure footage is tightly controlled within Tesla, with employees only assigned specific clips, which may not show the full extent of the technology's shortcomings. Within the company, certain teams were created to monitor close calls with pedestrians, labeled the "trauma team," who were allowed special permissions to review critical footage.
As the company prepares for public robotaxi demonstrations, workers have observed that although FSD has potential, achieving full autonomy remains challenging. Many felt that the system could not consistently handle complex driving scenarios safely.
Despite Musk previously heralding Tesla's software as prepared to operate everywhere, it became clear that significant mapping and specific local data training were necessary for the successful demonstration of FSD in certain environments. With numerous software updates, some capabilities improved while others deteriorated, leading to discouraging results when monitoring driver intervention statistics.
While Tesla has begun expanding its robotaxi service to other cities, the actual rollout has not met Musk's ambitious forecasts, with reports of limited availability and inconsistent performance from the autonomous vehicles. Critics emphasize that Tesla's methodology and claims regarding safety lack transparency and rigor when compared to more established competitors, which adhere to stricter safety analyses and testing protocols.
