Waymos crash less than human drivers

Interpreting the Safety Numbers

• Commenters broadly agree the reported 83–84% reduction in airbag‑deploying crashes is impressive, but note:
  • Sample sizes (13 vs 78 estimated crashes) are small with wide error bars.
  • The change 84%→83% is seen as “essentially unchanged,” even if framed as “slightly worse.”
• Some worry the comparison methodology (“same roads”) is under‑explained; human benchmarks are city‑level estimates adjusted to match Waymo’s service area, not exact same segments and times.
• Others highlight Waymo’s own crash logs and collision reconstructions as a positive transparency step, and note most recorded crashes appear to be other vehicles rear‑ending stopped Waymos.

Environment, Routes, and Generalization

• Persistent concern: Waymo operates only in selected cities, mostly in good weather, no freeways (until recently) and with prior mapping and route control; these are “easier miles” than the full spectrum of human driving.
• Defenders counter that:
  • SF city driving is chaotic and not “easy mode.”
  • Map data is a prior; vehicles detect construction, closures and update maps.
  • Limiting operation to conditions where capability is proven is itself safety‑positive.
• Open question: how well the system generalizes to truly novel environments (different cities, severe weather, unusual events) without heavy pre‑work.

Fault, Behavior, and Non‑crash Impacts

• Several argue that “crash count” alone may miss:
  • Near‑misses and confusion (e.g., odd behaviors at intersections, blocking traffic, looping roundabouts).
  • Crashes caused indirectly by AV behavior (e.g., overly cautious braking leading to human rear‑end collisions) even when legal fault lies with humans.
• Others maintain that, absent data, crashes per mile and severity remain the primary safety metric, while acknowledging more nuanced metrics (property damage, pedestrian impacts) would be useful.

Human Drivers, Distribution of Risk, and Regulation

• Multiple comments stress that crashes are highly skewed:
  • Roughly “20% of drivers cause 80% of serious crashes”; some data cited with drivers having dozens of near‑crashes in <20k miles.
• Proposals:
  • Stricter licensing, periodic retesting, and more serious DUI penalties.
  • Even banning or heavily restricting the worst drivers, with some suggesting AVs as a mandated alternative for high‑risk groups.
• Pushback:
  • In car‑dependent US environments, aggressive license revocation is seen as economically devastating and politically untenable.
  • Equity concerns: stricter testing and enforcement could be framed as discriminatory or de facto “driving for the rich only.”

Systemic Risks and Correlated Failures

• A key worry: correlated failure across a homogeneous fleet (e.g., bad software update, novel environmental shift, cyberattack) could cause rare but catastrophic multi‑car incidents, outweighing incremental lives saved.
• Mitigations discussed:
  • Staged rollouts of new software to subsets of the fleet.
  • Enabling new policies first on unoccupied trips.
• Some compare this risk profile to mass public‑health systems: low average risk but potentially large, rare tail events.

Economics, Pricing, and Business Model

• Experiences vary:
  • Some users report Waymo slightly cheaper than Uber/Lyft (especially with no tipping); others see it as consistently more expensive or similar but with longer wait times and slower routes (no highways, strict speed limits).
• Many doubt current economics:
  • High capex for specialized EVs, sensors, mapping, data centers, and human support staff.
  • Waymo reportedly still burning significant cash; question whether rides can become much cheaper than human‑driven services.
• Others argue that at scale, software and fleet centralization should beat the labor cost of millions of individual drivers, but acknowledge that today’s prices mostly reflect demand and experimentation, not final unit economics.

Autonomy Approaches: Waymo vs Tesla

• Commenters repeatedly distinguish:
  • Waymo: lidar + cameras, heavy mapping, geofenced service, no user control, strict reporting, small but real driverless fleet.
  • Tesla: vision‑only, owner‑driven everywhere, FSD as supervised assistance; many see it as impressive driver‑assist but far from safe unsupervised robotaxis.
• Debates:
  • Whether lidar is essential or a “crutch”; some see Tesla’s refusal to use lidar as ideology and cost‑driven, others as a legitimate long‑term bet.
  • Reliability in adverse conditions (night, fog, heavy rain); anecdotal examples where vision‑only systems misjudge distances or signals.

Urban Design, Transit, and Broader Impacts

• Strong thread arguing that:
  • Buses, trains, and cycling (with good infrastructure) are already safer per mile and healthier.
  • AVs risk entrenching car‑centric urban form instead of supporting dense, walkable cities.
• Counter‑arguments:
  • US transit construction costs and politics make large‑scale rail expansion extremely hard; AVs may be the most realistic near‑term improvement.
  • AV fleets could increase road capacity, reduce parking needs, enable smaller vehicles, and over decades reshape cities to be more human‑friendly.
• Many see AVs as complementary to transit (first/last mile), not a full substitute.

Public Perception, Metrics, and Adoption Path

• Several note that “better than average human” is a low bar; a more relevant benchmark might be experienced, sober, attentive drivers or professional drivers.
• However, because average driver performance includes drunk, distracted, and inexperienced drivers, replacing some of that population with safer AVs is still seen as a net win.
• Widespread belief: full replacement of human driving will be gradual and conditional on:
  • Demonstrably lower crash and fatality rates in diverse conditions.
  • Clear responsibility/liability frameworks.
  • Economic viability and user acceptance, including comfort with more cautious, rule‑bounded driving styles.