Uber Advanced Technologies Group test vehicle strikes pedestrian, Tempe, Arizona, 18 March 2018

Changes to what this case says, with the reason. Every earlier version is kept.

• 2026-10-21: Inference: signed by. The interpretation's wording changed after the editor signed it (second, independent check of 21 october 2026); the signature is cleared so the editor signs the corrected wording.
  Show the change

  Before: Saurabh Bhatnagar

  After:
• 2026-10-21: Inference: value. Found in the editor's second, independent check of the interpretations, 21 october 2026: abbreviation spelled out.
  Show the change

  Before: The system detected the pedestrian 5.6 seconds before impact but never classified her as a pedestrian or correctly predicted her path. When it identified the emergency 1.2 seconds out, avoiding her needed harder braking than its design allowed, so it held off braking for one second and then, as designed, did not brake hard to lessen the impact; the design relied on an operator who was visually distracted by her phone. In Volvo's simulation, which assumed ideal sensors, the factory emergency braking that had been switched off under computer control avoided the collision in 17 of 20 variations and cut impact speed below 10 mph in the rest, and the National Transportation Safety Board wrote that the tests suggest a standard Volvo XC90 with its factory collision warning and emergency braking might have prevented or at least mitigated the crash. The Board named the operator's distraction as the probable cause and found that the braking design increased risk, without listing it as a cause; the record does not show what the system's own braking from 1.2 seconds would have achieved

  After: The system detected the pedestrian 5.6 seconds before impact but never classified her as a pedestrian or correctly predicted her path. When it identified the emergency 1.2 seconds out, avoiding her needed harder braking than its design allowed, so it held off braking for one second and then, as designed, did not brake hard to lessen the impact; the design relied on an operator who was visually distracted by her phone. In Volvo's simulation, which assumed ideal sensors, the factory emergency braking that had been switched off under computer control avoided the collision in 17 of 20 variations and cut impact speed below 10 miles per hour in the rest, and the National Transportation Safety Board wrote that the tests suggest a standard Volvo XC90 with its factory collision warning and emergency braking might have prevented or at least mitigated the crash. The Board named the operator's distraction as the probable cause and found that the braking design increased risk, without listing it as a cause; the record does not show what the system's own braking from 1.2 seconds would have achieved
• 2026-10-21: Inference: value. Found in the editor's second, independent check of the interpretations, 21 october 2026: the board's statement concerns a standard volvo xc90 with its factory systems (ntsb-fin, page 40).
  Show the change

  Before: The system detected the pedestrian 5.6 seconds before impact but never classified her as a pedestrian or correctly predicted her path. When it identified the emergency 1.2 seconds out, avoiding her needed harder braking than its design allowed, so it held off braking for one second and then, as designed, did not brake hard to lessen the impact; the design relied on an operator who was visually distracted by her phone. In Volvo's simulation, which assumed ideal sensors, the factory emergency braking that had been switched off under computer control avoided the collision in 17 of 20 variations and cut impact speed below 10 mph in the rest, and the National Transportation Safety Board wrote that the tests suggest it might have prevented or at least mitigated the crash. The Board named the operator's distraction as the probable cause and found that the braking design increased risk, without listing it as a cause; the record does not show what the system's own braking from 1.2 seconds would have achieved

  After: The system detected the pedestrian 5.6 seconds before impact but never classified her as a pedestrian or correctly predicted her path. When it identified the emergency 1.2 seconds out, avoiding her needed harder braking than its design allowed, so it held off braking for one second and then, as designed, did not brake hard to lessen the impact; the design relied on an operator who was visually distracted by her phone. In Volvo's simulation, which assumed ideal sensors, the factory emergency braking that had been switched off under computer control avoided the collision in 17 of 20 variations and cut impact speed below 10 mph in the rest, and the National Transportation Safety Board wrote that the tests suggest a standard Volvo XC90 with its factory collision warning and emergency braking might have prevented or at least mitigated the crash. The Board named the operator's distraction as the probable cause and found that the braking design increased risk, without listing it as a cause; the record does not show what the system's own braking from 1.2 seconds would have achieved
• 2026-10-21: Inference: value. Found in the editor's second, independent check of the interpretations, 21 october 2026: the system did predict paths, and at 1.2 seconds it placed her on the vehicle's path (ntsb-fin, page 16).
  Show the change

  Before: The system detected the pedestrian 5.6 seconds before impact but never classified her as a pedestrian or predicted her path. When it identified the emergency 1.2 seconds out, avoiding her needed harder braking than its design allowed, so it held off braking for one second and then, as designed, did not brake hard to lessen the impact; the design relied on an operator who was visually distracted by her phone. In Volvo's simulation, which assumed ideal sensors, the factory emergency braking that had been switched off under computer control avoided the collision in 17 of 20 variations and cut impact speed below 10 mph in the rest, and the National Transportation Safety Board wrote that the tests suggest it might have prevented or at least mitigated the crash. The Board named the operator's distraction as the probable cause and found that the braking design increased risk, without listing it as a cause; the record does not show what the system's own braking from 1.2 seconds would have achieved

  After: The system detected the pedestrian 5.6 seconds before impact but never classified her as a pedestrian or correctly predicted her path. When it identified the emergency 1.2 seconds out, avoiding her needed harder braking than its design allowed, so it held off braking for one second and then, as designed, did not brake hard to lessen the impact; the design relied on an operator who was visually distracted by her phone. In Volvo's simulation, which assumed ideal sensors, the factory emergency braking that had been switched off under computer control avoided the collision in 17 of 20 variations and cut impact speed below 10 mph in the rest, and the National Transportation Safety Board wrote that the tests suggest it might have prevented or at least mitigated the crash. The Board named the operator's distraction as the probable cause and found that the braking design increased risk, without listing it as a cause; the record does not show what the system's own braking from 1.2 seconds would have achieved
• 2026-10-21: Controls: c-i4. Editor's classification call of 21 october 2026 (call 7, accepted from pull request #74): an auditory alert was designed and did fire, 0.2 seconds before impact, too late to act on; the control existed and failed rather than being absent.
  Show the change

  Before: absent

  After: failed
• 2026-10-20: Sources: ntsb-fin, sources: ntsb-pre, sources: az-plea (checked). Found in the editor's source re-check of 19-20 october 2026: each record re-read in full or in the relevant sections; titles, dates, types and links hold (the final report pdf served is the corrected copy reissued 26 june 2020, with corrections to the title page only).
  Show the change

  Before: 2026-10-13

  After: 2026-10-20
• 2026-10-20: Sources: az-susp. Found in the editor's source re-check of 19-20 october 2026: the cbs news page's exact headline is 'arizona gov. doug ducey suspends uber from autonomous testing' and it is credited cbs/ap ('the associated press contributed to this report').
  Show the change

  Before: {"title": "CBS News, 'Arizona governor suspends Uber autonomous testing', 26 March 2018 (reports the governor's letter)", "checked": "2026-10-13"}

  After: {"title": "CBS News/AP, 'Arizona Gov. Doug Ducey suspends Uber from autonomous testing', 26 March 2018 (reports the governor's letter)", "checked": "2026-10-20"}
• 2026-10-20: Sources: yav-ap. Found in the editor's source re-check of 19-20 october 2026: the washington times copy of the ap story could not be opened (it refused the fetch twice); the same ap story was read at arizona public media, and the source now points to that readable copy with its exact headline.
  Show the change

  Before: {"title": "Associated Press, 'Yavapai County Attorney: Uber not criminally liable' (Washington Times), 5 March 2019", "url": "https://www.washingtontimes.com/news/2019/mar/5/yavapai-county-attorney-uber-not-criminally-liable/", "checked": "2026-10-13"}

  After: {"title": "Associated Press, 'Arizona Prosecutor Says Uber Not Criminally Liable in Deadly Crash' (Arizona Public Media copy, updated 6 March 2019), 5 March 2019", "url": "https://azpm.org/s/64843-arizona-prosecutor-says-uber-not-criminally-liable-in-deadly-crash", "checked": "2026-10-20"}
• 2026-10-20: Inference: value. Found in the editor's source re-check of 19-20 october 2026: the ntsb final report shows the system kept braking up to 0.71 g (section 1.5.5.3), so 'removing the vehicle's authority to brake' overstated the design; the ntsb named the operator's distraction as the probable cause and found the braking design 'increased the safety risks' without listing it as a cause (p. v, p. vii, p. 40), so 'turned an uncertain detection into a fatality' was causation the record does not make; the 1.2-second figure and the only counterfactual the record makes (volvo's simulation, pp. 21-22 and 40) are in the final report, not the cited preliminary report. re-cited to ntsb-fin. the interpretation was not signed (signed_by already null); the editor must sign the new wording.
  Show the change

  Before: {"value": "Removing the vehicle's authority to brake turned an uncertain detection into a fatality. Had braking been enabled at 1.2 seconds, the outcome is uncertain, but it was not foreclosed.", "source_ids": ["NTSB-PRE"]}

  After: {"value": "The system detected the pedestrian 5.6 seconds before impact but never classified her as a pedestrian or predicted her path. When it identified the emergency 1.2 seconds out, avoiding her needed harder braking than its design allowed, so it held off braking for one second and then, as designed, did not brake hard to lessen the impact; the design relied on an operator who was visually distracted by her phone. In Volvo's simulation, which assumed ideal sensors, the factory emergency braking that had been switched off under computer control avoided the collision in 17 of 20 variations and cut impact speed below 10 mph in the rest, and the National Transportation Safety Board wrote that the tests suggest it might have prevented or at least mitigated the crash. The Board named the operator's distraction as the probable cause and found that the braking design increased risk, without listing it as a cause; the record does not show what the system's own braking from 1.2 seconds would have achieved.", "source_ids": ["NTSB-FIN"]}
• 2026-10-20: Controls: c-i4. Found in the editor's source re-check of 19-20 october 2026: the ntsb final report (p. 13 and timeline p. 16) shows no alert at the start of action suppression and an auditory alert at 0.2 seconds before impact; the preliminary report's 'not designed to alert the operator' was refined by the final report.
  Show the change

  Before: {"note": "Operator alerting on detected hazard", "confidence": "high"}

  After: {"note": "Operator alerting on detected hazard: no alert when the system detected the emergency 1.2 seconds before impact; an auditory alert came 0.2 seconds before impact", "confidence": "high"}
• 2026-10-20: Controls: c-r2. Found in the editor's source re-check of 19-20 october 2026: the ntsb final report (pp. 13-14, 22 and 41) shows braking up to 0.71 g was available; what was absent was harder braking, mitigation-only braking and the deactivated volvo automatic emergency braking.
  Show the change

  Before: {"note": "Automated emergency braking under computer control", "confidence": "high"}

  After: {"note": "Automated emergency braking under computer control: the system was not designed to brake beyond its 0.71 g cap or to brake hard only to lessen an unavoidable impact, and Volvo's automatic emergency braking was switched off", "confidence": "high"}
• 2026-10-20: Controls: c-r3. Found in the editor's source re-check of 19-20 october 2026: the ntsb final report calls the operator 'the primary countermeasure in an emergency situation' (p. 14), not the sole safeguard, since the system could brake within its 0.71 g cap; action suppression allowed the operator to take control (p. 40) with no alert (p. 13); atg consolidated to one operator in september to october 2017 (p. 29).
  Show the change

  Before: {"note": "Human supervision as sole safeguard; Second operator in vehicle (absent)", "confidence": "high"}

  After: {"note": "Operator takeover as the primary emergency safeguard: the one-second braking suppression allowed time for the operator to take control, but no alert was given when it began. Second operator in vehicle (absent: Uber moved to a single operator in September to October 2017)", "confidence": "high"}
• 2026-10-20: Fields: recovery accountability (settlement). Found in the editor's source re-check of 19-20 october 2026: npr's report of 29 march 2018 says the family's attorney confirmed a settlement between uber and the pedestrian's daughter and husband and gave few details; no primary record of the settlement was found, so it is recorded as the lawyer's assertion via press.
  Show the change

  Before:

  After: {"layer": "Recovery", "facet": "accountability", "value": "The family's lawyer said Uber had reached a settlement with the pedestrian's daughter and husband ('The matter has been resolved'); no terms were given (late March 2018, as reported by NPR on 29 March 2018)", "assertion_type": "assertion", "source_ids": ["SETTLE-NPR"], "confidence": "medium", "disputed": false}
• 2026-10-20: Fields: recovery accountability (yavapai review). Found in the editor's source re-check of 19-20 october 2026: the ap story of 5 march 2019 (arizona public media copy) says the case went to yavapai county after the metro phoenix prosecutor's office cited a potential conflict of interest, and that the letter returned the case to maricopa county prosecutors to determine whether to file criminal charges; ktar's report of the same day quotes the letter: 'no basis for criminal liability for the uber corporation'.
  Show the change

  Before: The Yavapai County Attorney, reviewing the case, found no basis for criminal liability for Uber (March 2019, as reported by AP)

  After: The Yavapai County Attorney, asked to review the case after the Maricopa County Attorney's Office cited a potential conflict of interest, found no basis for criminal liability for Uber and returned the case to the Maricopa County Attorney to decide whether to file criminal charges (5 March 2019, as reported by AP)
• 2026-10-20: Fields: recovery accountability (plea). Found in the editor's source re-check of 19-20 october 2026: the maricopa county attorney's office release of 28 july 2023 gives the plea date, the offense as 'endangerment, a class 6 undesignated offense', and that it would be designated a misdemeanor only on successful completion of the sentence.
  Show the change

  Before: The operator pleaded guilty to endangerment in July 2023 and was sentenced to three years of supervised probation

  After: The operator pleaded guilty on 28 July 2023 to endangerment, a class 6 undesignated offense, and was sentenced to three years of supervised probation; the judge determined the charge would be designated a misdemeanor only on successful completion of her sentence
• 2026-10-20: Fields: recovery technical (postcrash changes). Found in the editor's source re-check of 19-20 october 2026: the ntsb final report (section 1.9, pp. 29-31, and the postcrash-changes finding, p. vii) documents the company's changes after the crash: volvo systems re-enabled, action suppression removed, mitigation braking added, path prediction changed, two operators, and the limited pittsburgh testing. the case had no record of the fixes shipped.
  Show the change

  Before:

  After: {"layer": "Recovery", "facet": "technical", "value": "Uber's Advanced Technologies Group stopped public-road testing on 19 March 2018. When it resumed on 20 December 2018, Volvo's collision warning and automatic emergency braking stayed active under computer control; the system no longer suppressed braking after detecting an emergency and would brake even only to lessen a crash; it kept an object's tracking history across reclassifications and could predict a pedestrian crossing outside a crosswalk; each vehicle carried two operators; and testing was limited to a 1-mile loop in Pittsburgh at up to 25 mph (as of the NTSB final report, November 2019)", "assertion_type": "fact", "source_ids": ["NTSB-FIN"], "confidence": "high", "disputed": false}
• 2026-10-20: Fields: recovery automated fallback. Found in the editor's source re-check of 19-20 october 2026: the ntsb final report (executive summary p. v; section 1.5.5.3 p. 14) shows the system could brake up to 0.71 g where that avoided a collision and, after suppression, alerted the operator and initiated a gradual slowdown; 'none available' and 'relied entirely on the human' overstated the design. the register's reading is kept as an interpretation.
  Show the change

  Before: {"value": "None available: the design relied entirely on the human", "source_ids": ["NTSB-PRE"]}

  After: {"value": "No automated fallback once a collision could not be avoided within the system's braking limits: the design precluded emergency braking to lessen a collision and relied on the operator to avoid it or mitigate the impact; after the one-second suppression the system sounded an alert and initiated a plan for a gradual slowdown", "source_ids": ["NTSB-FIN"]}
• 2026-10-20: Fields: evidence actor. Found in the editor's source re-check of 19-20 october 2026: the ntsb final report (sections 1.5.6 and 1.5.6.2, p. 14 and p. 17): 'atg provided ntsb investigators with a comprehensive dataset', including the ads cameras and the janus v3 fleet-monitoring cameras; the data were provided by the company, not recovered by investigators.
  Show the change

  Before: {"value": "System logs and dashcam video recovered by investigators", "source_ids": ["NTSB-PRE"]}

  After: {"value": "Automated driving system data and video from the system's cameras and from an aftermarket camera unit facing forward, rearward and at the operator, provided by Uber's Advanced Technologies Group to the National Transportation Safety Board's investigators", "source_ids": ["NTSB-FIN"]}
• 2026-10-20: Fields: envelope authority. Found in the editor's source re-check of 19-20 october 2026: the ntsb final report (section 1.5.5.3, pp. 13-14) shows the system kept braking authority up to 0.71 g; what the design precluded was harder braking, any braking during the one-second action suppression, and braking only to mitigate an unavoidable crash. 'not permitted to execute emergency braking' was the preliminary report's summary and overstated the design.
  Show the change

  Before: {"value": "Under computer control the system was not permitted to execute emergency braking, by Uber's design", "source_ids": ["NTSB-PRE", "NTSB-FIN"]}

  After: {"value": "By Uber's design, automated braking under computer control was capped at 0.71 g (0.71 times the acceleration of gravity, 7 meters per second squared). When avoiding a collision needed harder braking than its limits allowed, the system suppressed braking for one second, and if the collision still could not be avoided within the cap it was not designed to brake hard just to lessen the impact", "source_ids": ["NTSB-FIN"]}
• 2026-10-20: Fields: decision. Found in the editor's source re-check of 19-20 october 2026: the ntsb final report (table pp. 15-16 and p. 16) shows the first classification was vehicle (by radar), then alternation among other, vehicle and bicycle, with tracking history dropped at each change; the order previously given was the preliminary report's, cited to the final report.
  Show the change

  Before: The software classified the pedestrian as an unknown object, a vehicle, then a bicycle, with changing path predictions; 1.2 seconds before impact it determined a collision was imminent

  After: The software never classified the pedestrian as a pedestrian: it classified her first as a vehicle, then alternated among unknown object, vehicle and bicycle, and at each change it dropped her tracking history, so it did not correctly predict her path. 1.2 seconds before impact it determined that her location was in the vehicle's path and a collision was imminent
• 2026-10-20: Fields: inputs. Found in the editor's source re-check of 19-20 october 2026: the ntsb final report's timeline table (p. 15) gives radar first detection at 5.6 seconds and 44.3 mph and lidar at 5.2 seconds; the preliminary report's 'about 6 seconds' and 43 mph were superseded by the final report.
  Show the change

  Before: {"value": "Radar and lidar registered the pedestrian about six seconds before impact at 43 mph", "source_ids": ["NTSB-PRE"]}

  After: {"value": "Radar first detected the pedestrian 5.6 seconds before impact, with the vehicle at 44.3 mph; lidar (light detection and ranging) detected her 5.2 seconds before impact", "source_ids": ["NTSB-FIN"]}
• 2026-10-20: Context: autonomy level text. Found in the editor's source re-check of 19-20 october 2026: the ntsb final report (sections 1.4.5.3 and 1.10.2) shows the operator requirement came from uber's operating procedures; arizona's executive order 2015-09 permitted testing whether or not a person was inside the vehicle.
  Show the change

  Before: Machine acts, human supervises (safety driver required)

  After: Machine acts, human supervises (safety driver required by Uber's operating procedures)
• 2026-10-20: Context: mandate context text. Found in the editor's source re-check of 19-20 october 2026: the ntsb preliminary and final reports both describe the system as a developmental automated driving system (final report title: 'vehicle controlled by developmental automated driving system'); no source calls it production testing.
  Show the change

  Before: Production testing on public roads

  After: Developmental testing on public roads
• 2026-10-13: Fields. Independent check against the linked records before right-of-reply letters.
  Show the change

  Before: 'not designed to alert the operator' and 1.3 s (preliminary report); 'no braking by the system'; probable cause given in part

  After: final report's findings: no alert during action suppression, alert 0.2 s before impact, 1.2 s; no emergency braking; full probable cause and contributing factors
• 2026-10-13: Fields and sources. Claims aligned to what each linked record says, found on review.
  Show the change

  Before: plea, non-prosecution and suspension cited to one AP-sourced record; NTSB reports cited 'as reported by' press; impact 'roughly 39 mph'

  After: plea cited to the Maricopa County Attorney's release; non-prosecution and suspension cited separately, typed as press-reported assertions; NTSB preliminary and final reports linked directly; 39 mph
• 2026-10-13: Context: jurisdiction. Jurisdiction text was split at a comma inside parentheses.
  Show the change

  Before: United States (Arizona | federal NTSB)

  After: United States (Arizona, federal NTSB)
• 2026-10-13: Controls: c-r3. A control has one state per case; entries in two states merged into the state the record supports.
  Show the change

  Before: failed: Human supervision as sole safeguard; absent: Second operator in vehicle

  After: failed: Human supervision as sole safeguard; Second operator in vehicle (absent)