Amazon’s self-driving robotaxi division, Zoox, has recalled 332 vehicles due to a critical software flaw causing them to unexpectedly drift into oncoming traffic. The recall, mandated by the National Highway Traffic Safety Administration (NHTSA), comes as public trust in autonomous vehicle safety remains low—with recent AAA data showing only 13% of people feel safe riding in them.
The Problem: Drifting Into Danger
The issue first surfaced on August 26 when a Zoox robotaxi made a wide right turn, swerving into an oncoming lane before briefly stopping there. Subsequent analysis revealed 62 similar incidents between August 26 and December 5. In these cases, the vehicles either crossed lane lines unnecessarily, especially near intersections, or stopped directly in the path of oncoming traffic. This behavior significantly increases collision risk, even though no accidents were reported prior to the recall.
Root Causes & Software Fixes
Zoox traced the bug to multiple factors:
– Misidentified parked cars (incorrectly labeled as moving obstacles)
– Unexpected route recalculations
– Flawed attempts to avoid blocking cross-traffic at intersections.
The company released an initial software update on November 7, followed by another earlier this month. As of December 19, all Zoox vehicles operating on public roads were updated with the latest fixes. However, Zoox maintains that it will continue monitoring vehicle performance and refining the software to prevent future occurrences.
Why This Matters
This recall highlights the ongoing challenges of deploying fully autonomous systems in real-world conditions. Even minor software errors can have severe consequences in driving scenarios, demonstrating that self-driving technology is not yet foolproof. The incident underscores the need for rigorous testing, redundancy, and continuous over-the-air updates to ensure safety before widespread public adoption.
While no collisions occurred, this incident reinforces that autonomous vehicles still require extensive refinement and oversight before they can be fully trusted to operate without human intervention.
