Waymo Issues Software Recall for Robotaxis Over School Bus Safety Issue
Waymo confronts a critical vulnerability in its autonomous driving software after incidents where robotaxis illegally passed stopped school buses unloading children. The Alphabet subsidiary’s fifth-generation system misinterpreted extended stop signs and flashing red lights, prompting a voluntary recall to federal regulators. This development arrives as robotaxi fleets expand across U.S. cities, raising questions about software reliability in high-stakes pedestrian zones.
The recall targets all vehicles equipped with Waymo’s fifth-generation autonomous hardware, numbering over 1,000 Jaguar I-Pace units operating in San Francisco, Los Angeles, Phoenix, and Austin. Software logs revealed 19 such passings in Austin alone during 2025, with at least five occurring post a November 17 update designed to enhance signal detection. Federal investigators at the National Highway Traffic Safety Administration initiated scrutiny following a viral video from Atlanta depicting a Waymo vehicle crossing in front of a bus with active warning signals.
The core issue stems from the system’s perception algorithms, which integrate lidar, radar, and high-resolution cameras to classify dynamic road elements. In these cases, the software initially decelerated but failed to sustain a full stop, overriding precautionary thresholds due to incomplete mapping of school bus geometries in edge-case scenarios. No injuries resulted from the documented events, but the episodes violated state laws mandating 25-foot buffers around active school buses, potentially exposing young passengers to rear-end collision risks from trailing traffic.
Waymo filed the recall notice with NHTSA on December 8, deploying an over-the-air patch that recalibrates the decision-making pipeline for 98% confidence in bus signal prioritization. The update incorporates expanded training data from 500 simulated crossings, reducing false negatives by 75% in internal benchmarks. Deployment across the fleet requires 24 hours of validation per vehicle, with full rollout completing by December 15. Regulators have 30 days to review the submission, focusing on post-fix telemetry to confirm compliance with Federal Motor Vehicle Safety Standard 108.
Mauricio Peña, Waymo’s chief safety officer, addressed the filing in a company blog: “Holding the highest safety standards means recognizing when our behavior should be better. As a result, we have made the decision to file a voluntary software recall with NHTSA related to appropriately slowing and stopping in these scenarios.” Waymo emphasized its overall safety metrics, citing 12 times fewer pedestrian injury crashes than human drivers per million miles driven.
This recall underscores broader challenges in scaling Level 4 autonomy, where software must process 1.5 terabytes of sensor data per hour amid unpredictable urban variables like erratic school routes. Competitors including Cruise and Zoox report similar calibration hurdles, with NHTSA probes into 15 autonomous incidents nationwide in 2025. Waymo’s operations, spanning 300 square miles in four markets, logged 50 million autonomous miles by Q3, yet these lapses highlight the gap between aggregate statistics and scenario-specific robustness.
For U.S. regulators, the case accelerates calls for standardized testing protocols under the SELF DRIVE Act, mandating disclosures of software update efficacy within 72 hours. Waymo’s response aligns with industry norms, where over-the-air fixes resolve 85% of non-mechanical defects without physical inspections. Austin public school officials, who flagged the passings via dashcam submissions, welcomed the action but demanded quarterly audits through 2026.
The incident coincides with Waymo’s expansion plans, including a 20% fleet increase in Phoenix by Q1 2026 and entry into Atlanta’s downtown grid. Safety engineers note that while human drivers commit 2,000 school bus passings annually per AAA data, autonomous systems must exceed that baseline to gain public trust. This recall, though contained, reinforces the iterative nature of self-driving tech, where incremental updates bridge the divide between prototype and production-scale deployment.
