Skip to content
ENERGYDM Group
  • Home
  • Blog & Insights
  • Services
  • Reports
  • About
  • Contact
  • Member Area
Reports & Analyses
Log In

The Xiaomi SU7 Highway Crash: What Happened, Why It Matters, and What Comes Next?

Posted by:

ENERGYDM Group

|

On:

May 23, 2025

|

1  Accident Overview

On 29 March 2025, a base-trim Xiaomi SU7 travelling on the Dezhou–Shangrao Expressway in Anhui, China, struck a roadside concrete barrier while Navigate on Autopilot (NOA) was active. System logs show the car was cruising at 116 km h-1 when the NOA issued an obstacle alert; the driver grabbed the wheel one second later, but the sedan still hit the barrier at ≈97 km h-1, burst into flames, and all three occupants were killed [1]. Xiaomi told police it is co-operating fully and has already supplied raw driving data for download [2].

2  Crash Timeline and Key Telemetry

Local time (hh:mm:ss)EventVehicle state
22 : 27 : 17NOA engaged116 km h-1
22 : 28 : 17Mild-distraction warning—
22 : 36 : 48“Hands on wheel” alert—
22 : 44 : 24Obstacle ahead alertBrake request
22 : 44 : 25Driver takes controlWheel ≈ 22° L, brake 31 %
22 : 44 : 26Impact with barrier97 km h-1
22 : 44 : 28eCall triggered—
(The sequence is reconstructed from Xiaomi’s own event log [3] and highway-patrol data [1]).

The human had ≈0.8 s from alert to impact—far below the 2.3 s average NHTSA reaction benchmark. AEB logic slowed the car ~19 km h-1 but did not fully stop because the software treats rigid, stationary objects differently from moving lead vehicles [1].

3  Technical Factors

  • Sensor suite (base SU7) — 11 cameras, 3 mm-wave radars, 12 ultrasonics; no LiDAR [6].
  • Compute — Single NVIDIA DRIVE Orin-X SoC (254 TOPS) running Xiaomi Pilot (L2+) software.
  • Battery — 73.6 kWh BYD FinDreams LFP “blade” pack (400 V) [8].
  • Brakes & chassis — Brembo four-piston front calipers, Bosch ABS/ESC, ZF dampers [9] , [10] .

Why did AEB not save the car?  Xiaomi’s driver-assist logic prioritises moving obstacles; at highway speeds the camera–radar fusion often filters out static objects to avoid false positives. With only vision and radar, the barrier’s geometry may have been mis-classified until it filled the field of view, leaving too little braking distance [13].

4  Camera-Only vs LiDAR Debate

Higher SU7 trims (Pro/Max) add a roof-mounted Hesai AT128 LiDAR and a second Orin for 508 TOPS of compute [7]. LiDAR gives centimetre-grade range and detects low-contrast obstacles—even at night or in glare—that cameras struggle with. Tesla’s all-camera Autopilot faces similar scrutiny: the U.S. NHTSA is probing 2.4 million Teslas after multiple stationary-object crashes [11], while a 2023 recall addressed Autopilot misuse when hitting parked emergency vehicles [12]. Critics argue the Xiaomi accident shows why redundant sensing remains prudent until full-stack vision AI matures [13].

5  Global Safety & Validation Standards

  • China—after the crash, MIIT banned terms like “smart” or “autonomous driving” in ads and now requires type-approval for over-the-air (OTA) driving-software updates [17].
  • United States—NHTSA is adding four ADAS test items to NCAP and is openly investigating Tesla FSD performance in low-visibility scenarios [15], [11].
  • European Union—Euro NCAP’s 2024 protocol roughly doubles the number of assisted-driving scenarios and emphasises stationary-object AEB performance [14].

These tighter test matrices contrast with China’s rapid “release-first, refine-later” cycle, where new entrants like Xiaomi shipped the SU7 in barely two years.

6  Market Fallout

Within two weeks of the crash, new SU7 orders fell ≈55 % month-on-month as refunds and social-media backlash mounted [4]. Analysts fear the incident could slow Chinese EV expansion abroad just as the EU weighs tariffs on subsidised imports [18]. Meanwhile, LiDAR suppliers such as Hesai expect the tragedy to accelerate multi-sensor demand and are already planning cheaper next-gen units to reach mass-market price points [7].

7  How the SU7 Compares

MetricXiaomi SU7 (base)Tesla Model 3 (2025)NIO ET5EU Benchmark (Euro NCAP ‘24)
SensorsCamera + radarCamera onlyCamera + LiDAR + radarCamera + LiDAR + radar (test spec)
AEB stationary target (110 km h-1)Partial stop (software-limited)Often no brakingFull stopMust stop to score “Very Good”
Development cycle≈24 mo≈18 mo refresh≈36 mo48 – 60 mo typical
Recent fatal ADAS crash?Yes (2025)Yes (2023, 2024)Yes (2021)—

8  Take-Aways for Engineers & Enthusiasts

  • Redundancy matters—A single perception modality can miss edge-cases. Multi-sensor stacks cost more but mitigate blind-spots.
  • Human factors remain critical—Even the quickest alerts are futile without enough reaction time; driver-monitoring must escalate earlier.
  • Validation is no longer optional—Euro NCAP’s 40 new tests and MIIT’s OTA-approval rule show regulators aligning safety with software pace.
  • Market perception can swing overnight—A single high-profile crash cut SU7 demand in half and sparked nationwide ad-language reform.

9  Conclusion

The March 29 Xiaomi SU7 tragedy is a stark reminder that today’s Level-2 “pilot” features are assistance, not autonomy. Vision-heavy systems can work brilliantly in most conditions, but the physics of a 2-tonne sedan give drivers and algorithms mere fractions of a second to react when things go wrong. As China tightens oversight and Europe and the U.S. expand ADAS testing, EV makers—new and established—must balance speed-to-market with exhaustive safety validation. Until then, the smartest car on the road is still the one with an alert human behind the wheel.

References

[1] CarNewsChina, “First fatal accident involving Xiaomi SU7 claims three lives,” 1 Apr 2025.

[2] Reuters, “China’s Xiaomi says it is cooperating with police after fatal EV accident,” 1 Apr 2025.

[3] 36氪 (36kr), “Behind the Xiaomi car accident: intelligent driving under scrutiny,” 2 Apr 2025.

[4] Reuters, “Xiaomi’s new EV orders slump in China as consumer backlash grows,” 14 May 2025.

[5] Reuters, “China bans ‘smart’ and ‘autonomous’ driving terms from vehicle ads,” 17 Apr 2025.

[6] “Xiaomi SU7,” Wikipedia, accessed 19 May 2025.

[7] Hesai Technology, “Xiaomi Unveils Revolutionary SU7… Equipped with AT128 LiDAR,” press release, 2 Apr 2024.

[8] CNEVPost, “Xiaomi SU7 standard gets random CATL/BYD batteries,” 27 May 2024.

[9] SMBOM.com, “Who are the key suppliers for Xiaomi SU7?” Jan 2025.

[10] Brembo, “Precision and power for the Xiaomi SU7 Max,” product page, Apr 2025.

[11] Reuters, “US probes Tesla’s Full Self-Driving software in 2.4 million cars,” 18 Oct 2024.

[12] Reuters, “Tesla recalls over 2 million vehicles after Autopilot crashes into emergency vehicles,” 13 Dec 2023.

[13] Fast Company, “Why Tesla’s camera-only approach may be a mistake,” 19 Mar 2025.

[14] AB Dynamics, “What the 2024 update to Euro NCAP’s assisted-driving grading means,” 2024.

[15] NHTSA, “NCAP Final Decision Notice—ADAS Roadmap,” Nov 2024.

[16] CarNewsChina, “Xiaomi SU7 scores five stars in C-NCAP crash test,” 25 Mar 2025.

[17] Gizmodo, “China bans ‘autonomous driving’ claims from car marketing following crash,” 17 Apr 2025.

[18] Focus (CBBC), “China’s autonomous-driving crackdown: global implications,” 24 Apr 2025.

Posted by

ENERGYDM Group

  • ENERGYDM Group Selected by LEMAN Group for Strategic Advisory Engagement
    Free Insights
    October 1, 2026

    ENERGYDM Group Selected by LEMAN Group for Strategic Advisory Engagement

  • How close are we to a true software-defined vehicle?
    ADAS & Autonomous Driving · AI & Advanced Computing · Automotive Market & Strategy · Free Insights · Software-Defined Vehicles
    September 27, 2026

    How close are we to a true software-defined vehicle?

  • Why Electric Vehicles Still Need Specialized Fluids
    Electrification & Batteries · Free Insights
    September 18, 2026

    Why Electric Vehicles Still Need Specialized Fluids

  • Software Scalability as Part of the EV Affordability Equation
    Free Insights · Software-Defined Vehicles
    September 17, 2026

    Software Scalability as Part of the EV Affordability Equation

ENERGYDM Group

Automotive intelligence, strategic advisory and market influence.

Connect on LinkedIn →

Company

About
Services
Contact

Intelligence

Reports
Industry Insights Membership
Enterprise Intelligence
Blog & Insights

Legal

Privacy Policy
Terms & Conditions


© 2026 ENERGYDM Group. All rights reserved.

Your cart (items: 0)

Products in cart

Product Details Total
Subtotal $0.00
Taxes and discounts calculated at checkout.
View cart
Go to checkout

Your shopping cart is empty

Return to shop

Notifications

ENERGYDM Group

ENERGYDM GROUP INDUSTRY INSIGHTS

Turn automotive change into clearer decisions.

Access in-depth analysis on software-defined vehicles, ADAS, E/E architecture, electrification and automotive AI.

  • Complete premium report library
  • SDV, ADAS, AI and electrification coverage
  • New analysis throughout the year
  • One annual membership
Explore Industry Insights → Already a member? Log in
ENERGYDM Group Rivian Zonal E/E Architecture short report

COMPLIMENTARY SHORT REPORT

Not ready to join?

Preview our complimentary Rivian short report.

Preview the free report →
ENERGYDM Group

ENERGYDM GROUP INDUSTRY INSIGHTS

Turn automotive change into clearer decisions.

Access in-depth analysis on software-defined vehicles, ADAS, E/E architecture, electrification and automotive AI.

  • Complete premium report library
  • SDV, ADAS, AI and electrification coverage
  • New analysis throughout the year
  • One annual membership
Explore Industry Insights → Already a member? Log in
ENERGYDM Group Rivian Zonal E/E Architecture short report

COMPLIMENTARY SHORT REPORT

Not ready to join?

Preview our complimentary Rivian short report.

Preview the free report →