Notes on Cruise's Pedestrian Accident
A detailed post-mortem of Cruise’s October 2023 pedestrian collision in San Francisco raises concerns about both the company’s behavior and the adequacy of current investigative practices for autonomous vehicles. Commenters question why police released the AV so quickly, criticize Cruise’s selective video sharing and “connectivity issues,” and argue for aviation-style, transparent accident investigations with real accountability for operators and manufacturers. The incident also intensifies broader debates over when (or whether) fully driverless cars should be allowed on public roads, especially given risks to vulnerable groups like blind pedestrians and the industry’s incentives to prioritize rapid deployment over safety.
Police response and evidence handling
- Some are surprised the Cruise vehicle was returned to the company within ~4 hours; they note ordinary drivers’ cars are typically held longer after serious pedestrian collisions.
- Others argue many usual reasons for impound (identifying driver, drug/alcohol checks, fingerprints) don’t apply to AVs, so retaining the car may have little benefit.
- Several think protocols simply don’t exist yet for AV crashes, so police defaulted to “normal car crash” behavior and lacked tools to investigate software or data.
Cruise’s post-incident behavior and video sharing
- Repeated “internet connectivity issues” during regulatory briefings are widely viewed as suspicious, especially as they coincided with obscuring the dragging sequence.
- Some note investigators accepted low-quality screen-sharing instead of demanding original footage or clear, complete reports.
- There’s broad agreement that the apparent omission/obfuscation is more damaging than the underlying technical failure, undermining trust in Cruise and AV reporting.
Technical limitations: scene understanding
- Multiple comments focus on lack of “scene understanding”: the AV accelerated once the pedestrian left its lane, did not slow for a rule-breaking pedestrian, and did not instantly stop when the adjacent car hit them.
- People contrast this with what a cautious human driver might do: anticipate erratic pedestrians, react to others’ collisions, and adopt anomaly-based defensive driving.
- There is debate on whether current ML approaches can model complex, multi-agent traffic behavior and subtle cues, or whether this points to deeper architectural limits.
Safety, risk, and deployment strategy
- Some argue AVs should only operate without safety drivers once they are demonstrably much safer than humans (e.g., an order of magnitude), even if that delays deployment for years.
- Others counter that tens of thousands already die annually from human driving; if AVs are even modestly safer, delaying broad deployment costs lives.
- There is disagreement over whether available data shows Cruise/Waymo are “comparable” to humans or whether current stats are too narrow, vendor-controlled, or biased to support strong claims.
Impact on vulnerable pedestrians and service quality
- A blind pedestrian expresses fear that AVs increase personal risk and that legislators are trading safety of small vulnerable groups for “innovation.”
- Some respondents say that in a mature AV future, blind pedestrians would be far safer than with human drivers, but concede we are not at that stage.
- Concerns are raised about accessibility and service quality: human taxi drivers provide assistance (calling out arrival, guiding to doors) that robotaxis currently cannot match, even if they become cheaper.
Accountability, regulation, and comparisons
- Several participants argue AV crashes should be investigated like aviation or rail accidents: detailed, blame-light technical inquiries, but with strong sanctions for concealment or negligence.
- Others insist that blame and legal accountability are essential to motivate safety investments, as long as they target the right actors (companies, designers, road authorities) rather than low-level scapegoats.
- Comparisons to aviation and past industrial scandals (emissions cheating, safety shortcuts) are used to argue that carmakers and tech firms cannot simply be trusted; strong, specialized regulators are needed.
Fault, human behavior, and road design
- The initial collision is widely acknowledged to be caused by the human driver and/or pedestrian crossing against the signal; the AV’s fault lies in the dragging and inadequate situational response.
- Some contend most crashes involve multiple contributing failures (pedestrian behavior, drivers, road design, enforcement), not a single clear “villain.” Others say many cases do have one plainly at-fault driver.
- Suggestions include holding road designers and manufacturers to higher standards, redesigning infrastructure, and potentially treating serious road crashes with far more investigative rigor.