Human Tech Tree
Current research2020–2026 · Present (2015 – Oct 2026)

Energy & Industry / Transport & Mobility

Robotaxis in Daily Service

Waymo passed 500,000 paid rides a week in March 2026 and now serves 14 US cities; Baidu Apollo Go has passed 22 million rides across 27 cities.

Open in the interactive tree →

Robotaxis are autonomy level 4: the vehicle drives without a driver, but only in defined areas and conditions. Waymo (Alphabet) and Baidu lead; Tesla runs a very small unsupervised fleet in Texas.

As of October 2026

Waymo passed about 500,000 paid rides per week in March 2026 and now runs around 4,000 vehicles in 14 US cities (TechCrunch, 24 September 2026), and its own safety data to June 2026 show 95% fewer serious-injury crashes than human drivers over 271 million rider-only miles. Baidu Apollo Go had completed more than 22 million rides by April 2026 and operates in 27 cities; test drives in London began in July 2026, with public rides planned from 2027 subject to approval. Tesla ran only about 25 unsupervised robotaxis in Austin, Dallas and Houston in spring 2026 by observers' counts and ties large-scale unsupervised operation to FSD v15 (late 2026 or early 2027).

Open steps

  • Statistical safety case for rare events High AI leverageShow with confidence that a driver is safer than humans in rare, severe situations, using limited real miles plus simulation and fleet logs.
  • Launching a city with little local data High AI leverageGeneralize the driver to new cities and road layouts with few local test miles, as gains from more data and compute become predictable.
  • Snow, heavy rain and unmapped roads Medium AI leverageKeep perception and planning reliable when lane markings, sensors and maps fail; first true winter-city rollouts are only starting.
  • Fewer remote helpers, cheaper vehicles Medium AI leverageCut human remote-assistance calls per mile and the sensor and compute cost per vehicle, which set fleet economics.

Where AI could help

High AI leverage. Driving skill is learned software, so more data, compute and testing raise performance directly; vehicles, permits and city validation still pace growth.

  • Train driving models on fleet data, where scaling laws show gains from more compute and data
  • Generate rare scenarios in simulation to test safety before entering each new city
  • Detect and handle edge cases such as emergency vehicles and construction zones
  • Optimise fleet dispatch, charging and remote assistance

Shown so far

  • In June 2025, a Waymo study on 500,000 hours of driving data found that model performance improves as a power law of training compute, as seen in language models. source

Prerequisites

Unlocks

Sources

More in Transport & Mobility · Present

All 50 points in Transport & Mobility →

Open in the interactive tree →