Editorial Standards

Driverless Daily is published by GreenRobot LLC, which runs the site, the daily broadcast and the podcast, and is responsible for what they say. Write to info+driverless@greenrobot.com or use the contact page.

We are not affiliated with any vehicle maker or operator. No one pays to be covered, ranked or left out, and the show runs no sponsored segments.

Self-driving vehicles and the people building, testing and regulating them: robotaxis, driver assistance, autonomous trucking, safety and policy.

Each day the site gathers stories from publishers’ own feeds, operators’ blogs, Google News, Hacker News and Reddit, and keeps only those about self-driving technology. A local language model then acts as editor and picks the day’s stories. It favors real deployments and expansions with specifics, safety data and crash investigations, regulatory decisions and permits, genuine technical developments, and funding news that changes what a company can do, and it spreads its picks across outlets.

It is told to leave out stock-price commentary with no news in it, rumors and unsourced timelines, press releases restated without substance, and duplicates of a story already picked.

Every gathered story stays readable on the news page. Search engines are offered only the stories that aired in a broadcast or that several outlets covered; a single outlet’s story that never aired is a summary of someone else’s reporting, and that reporting is the better page to find.

• Summaries. A language model running on our own computer summarizes each story, in its own words, from what the outlet published; for some sources that is only a headline and a line. The story page links the original.
• The script. The same model writes the two-anchor script from those summaries alone. A checking pass then compares every line with its source summary and sends back any figure, date or quotation the summary does not contain, and any story that was not credited.
• The anchors. Lane and Otto are characters drawn as robots. Their voices are synthesized; no person reads the show.
• The pictures. The broadcast tries to give each story one AI illustration, marked “AI illustration”; if the picture step fails, the show airs without them. The image model is told to show a place, a vehicle or a piece of infrastructure, and never a person, a face, text or a logo, so that a picture reads as an illustration, not a photograph of what happened. No one checks each picture before it airs, so if one gets that wrong, tell us.

No person reviews each broadcast before it airs: the checks above are automated. That is why every story links its source, and why corrections are taken seriously. How the broadcast is made has the full detail.

• The anchors report what the source summary says and nothing it does not, so a company’s account of its technology, a crash or a launch stays that company’s account.
• The crash tracker counts the crash reports companies file with the U.S. safety regulator, NHTSA. It says on every page that those reports may be incomplete or unverified, and that being in a crash is not causing it.

When a broadcast or a story page is wrong, we correct it and say what changed and when, on the page itself. The corrections policy has the details. Tell us through the contact page; you do not need an account.