HN Debrief

Tesla Full Self Driving uses bicycle lane in official Denmark approval video

  • AI
  • Transportation
  • Regulation
  • Europe

The story centers on Tesla’s official Europe promo video for Full Self Driving in Denmark. In the clip, the car turns right where that maneuver is illegal and starts entering a bike-only street marked with signs and lane markings that should have blocked the move. That landed badly because this was not a random owner video. It was curated marketing material released while Tesla is expanding approvals in Europe, so people read it as a glimpse of what the company itself thinks is acceptable evidence of readiness.

If you build or buy autonomy, assume edge cases are really jurisdiction cases. Local traffic rules, signs, and liability regimes are product requirements, not cleanup work after launch.

Discussion mood

Strongly negative. People saw the video as embarrassing for Tesla because it came from Tesla itself, showed a basic illegal maneuver around cyclists, and reinforced a broader view that the company is pushing autonomy before it has nailed signage, local rules, and accountability.

Key insights

  1. 01

    Local traffic law is a core autonomy problem

    What looks like a simple bad turn exposes a scaling problem. Traffic rules differ not just by country but by state and city, including whether you must merge into a bike lane before turning or must never enter it at all. An autonomy stack cannot treat those as cosmetic map details. It needs explicit jurisdiction-aware behavior or it will confidently do the wrong thing in places that look visually familiar.

    If your product operates on public infrastructure, build policy and legal variation into the architecture early. Expansion into a new market is not just more data collection. It is a fresh compliance surface.

      Attribution:
    • tzs #1
    • rootusrootus #1
  2. 02

    The issue was the turn itself, not lane drift

    The clip did not just show a sloppy encroachment into a bike lane. Commenters noted that turning right there was illegal in the first place, and the lane it entered was a bike-only road with clear markings and opposing traffic markers. That changes the reading from "imperfect control" to "the system chose a forbidden path."

    Separate control errors from policy errors when you evaluate autonomy. A system that picks an illegal maneuver needs different fixes and a different safety case than one that merely executes a legal plan poorly.

      Attribution:
    • sburud #1
    • chrononaut #1
    • ShinyLeftPad #1
  3. 03

    Learning from humans can import illegal behavior

    Training on real driving has a nasty failure mode. If local drivers routinely bend rules because road design nudges them that way, the model can learn those violations as normal behavior. One commenter pushed that further and said this even creates cover for a company to ship behavior that looks locally adapted while avoiding responsibility for having encoded rule breaking on purpose.

    Treat imitation learning from fleet behavior as contaminated data. You need hard constraints and audits for legally sensitive actions, especially around bike lanes, school zones, and signed exclusions.

      Attribution:
    • tzs #1
    • f33d5173 #1
  4. 04

    Hard-sign failures would be more damning than the turn

    Several comments argued the worst case is not a bad map or a Copenhagen edge case. It is failure to obey explicit signs like no-entry or one-way restrictions. One commenter claimed that if Tesla still cannot reliably recognize and prioritize standard traffic signs, that points to a product choice rather than an unsolved research problem, because sign classification is a relatively tractable vision task compared with full scene understanding.

    Ask vendors for evidence on simple rule-gating tasks, not just end-to-end demo miles. If they cannot prove strong performance on sign handling and route exclusion, broader autonomy claims are hard to trust.

      Attribution:
    • Veserv #1 #2
    • runarberg #1
  5. 05

    Weak basic driver assist undermines the FSD pitch

    Owners used the story to vent about Tesla’s ordinary assist features, especially phantom braking and lane centering that lags rivals. The interesting part was the business read. Some think Tesla has let Autosteer stagnate to push upgrades into FSD, but that strategy cuts against trust because bad baseline assistance makes customers less willing to believe the premium autonomy stack is safe.

    Do not expect a premium autonomous tier to outrun poor quality in the base product. Reliability in the everyday assist features is part of the credibility stack for anything more advanced.

      Attribution:
    • lnsru #1 #2
    • rootusrootus #1
    • speedgoose #1

Against the grain

  1. 01

    This could be a fixable mapping bug

    A minority view treated the clip as exactly the sort of error you expect during iterative rollout. A human would get ticketed, the navigation or perception stack would be updated, and the system would improve. That framing rejects the jump from one illegal turn to a blanket conclusion that the whole program is uniquely dangerous.

    Do not infer system-wide failure from a single vivid clip without checking error rates and failure classes. The right question is whether the vendor can identify, patch, and prevent recurrence quickly.

      Attribution:
    • red75prime #1 #2
  2. 02

    FSD may be much better than old Autopilot

    One pushback on owner complaints was that people are mixing Tesla’s aging basic driver-assist stack with the newer Full Self Driving stack. If true, anecdotes about old Autopilot lane keeping and cruise behavior say less about the specific software shown in the Denmark video than critics suggest.

    When comparing autonomous products, pin every claim to the exact software tier and version. Mixed anecdotes across product lines can make risk assessment noisier than it needs to be.

      Attribution:
    • vardump #1

In plain english

Autopilot ↩
Tesla’s broader name for its standard driver-assistance features such as adaptive cruise control and lane keeping.
Autosteer ↩
Tesla’s lane-centering driver-assistance feature that helps keep the car within a lane under supervision.
FSD ↩
Full Self Driving, Tesla’s advanced driver-assistance package.
Full Self Driving ↩
Tesla’s branded advanced driver-assistance software, often shortened to FSD, which aims to automate more of the driving task but still requires human supervision in many markets.
phantom braking ↩
A problem where a driver-assistance system suddenly brakes even though there is no real hazard ahead.

Reference links

Primary source video

Related driving footage and criticism