After years of struggling with unprotected left turns, Tesla’s Full Self-Driving system has made notable progress. The company first acknowledged these difficulties in 2022 through FSD v10.13 release notes. Since then, Tesla’s internal testing team has focused exclusively on improving performance in these challenging situations for over three years.
Unprotected left turns happen when a driver must cross oncoming traffic without a green arrow. They’re considered one of the most difficult problems for autonomous driving systems. Tesla identified at least 1,000 similar scenarios across the United States. A Jacksonville, Florida intersection near Memorial Park became the company’s primary testing location for this challenge.
Early versions of FSD struggled considerably with these turns. The system operated on a simple rule: both traffic lanes needed to be completely clear before proceeding. This approach didn’t work well in real traffic. The software also couldn’t judge small gaps safely or use medians as safety buffers. Sometimes it’d brake in the middle of a turn, creating dangerous situations.
Testing revealed performance problems in various conditions. Early versions hesitated when detecting vehicles in right lanes, interfering with traffic flow. Low-light situations like twilight created additional challenges. Beta testers like Chuck Cook documented multiple attempts at difficult intersections, sometimes testing during dawn hours to evaluate the system’s low-light capabilities. Despite technical success in controlled environments, user experiences have been described as terrifying due to uncertainty about the system’s decision-making reliability.
Progress has been steady. FSD v12.3 showed hesitation issues. Version 12.4.1 brought incremental improvements. Version 12.5.4.2 successfully handled five challenging unprotected left turns in New York City. The latest version, FSD v14.1.2, demonstrated what testers called “stunning” performance improvements on the historically difficult Jacksonville intersection.
The improvements suggest Tesla’s system is learning to handle unprotected left turns more like experienced human drivers. Modern versions can better judge traffic gaps and time their turns appropriately. The system no longer needs both lanes completely clear. Instead, it’s becoming more confident in evaluating safe turning opportunities. These advances are part of Tesla’s broader safety approach, where human error accounts for 90% of crashes and autonomous systems help address common driver mistakes.
Tesla’s work on this problem shows how specific challenges in autonomous driving require sustained focus. The company’s multi-year commitment to perfecting unprotected left turns has produced measurable results. As the system continues improving, these once-problematic maneuvers are becoming routine.