Can Xiaomi Out-Telsa Tesla?
The recent Xiaomi SU7 fatal crash is a sobering lesson for a maverick EV brand that has been moving fast and outselling Tesla’s Model 3 in China since December.
(Image: Xiaomi)
The Xioami SU7 sedan that killed three college students last month in Anhui Province is a stark reminder that automobile safety is a tough nut to crack. This is a sobering lesson — if heeded — for a maverick EV brand that has been moving fast and outselling Tesla’s Model 3 in China since December.
An industry observer, who spoke on condition of anonymity, quipped: “Sadly, this is what you get when you try to out-Tesla Tesla.”
Smartphone giant Xiaomi has been riding the electric vehicle wave in China, which has significantly lowered entry barriers for many non-traditional auto companies.
Coupled with “China speed”— applied to everything from goods delivery, design and manufacturing to technology innovation and fundamental lifestyle—Xiaomi, already a leading smartphone brand, moved into automotive last year.
Whether Xioami, with a 15-year record of software prowess on the smartphone market, can master the “intelligent driving” software stack in the automotive world poses an entirely different riddle.
What has been reported in China
First, here’s what we know.
Right now, there isn’t a whole lot of information on the accident, certainly not enough to assign blame for the deadly crash on software, hardware, the vehicle’s sensor suite or the driver.
Xiaomi has noted that it is fully cooperating with an investigation by the police in China.
Thus far, Xiaomi has disclosed the following:
· The car was in the “Navigate on Autopilot” (NOA) intelligent-assisted driving mode before the accident. It was moving at 116 kph (72 mph).
· The autopilot system had issued a warning of obstacles ahead.
· A driver took over and tried to slow down but the car collided with a cement pole at a speed of 97 kph.
The Chinese press and social media added this:
· Local traffic police told the father of one victim that the car caught fire after hitting the utility pole.
· It was alleged that after the vehicle caught fire, the “doors could not be unlocked, preventing escape.”
‘Been there, done that’
For many reporters covering Tesla’s numerous accidents in the United States, the details revealed about Xiaomi’s first fatal crash had a distinct “been-there, done-that” feel.
I don’t mean to be blasé. But issues like the difficulty of a handover from computer driver to human control, automation complacency on the part of the human driver, a post-crash fire triggered by EVs, vehicle doors that won’t unlock … none of these hazards is remotely unprecedented.
In the United States, despite the investigations of fatal crashes and harsh rebukes by the National Transportation Safety Board, followed by lawsuits, Tesla has thus far successfully blamed its deadly crashes on human drivers.
With its brand largely unscathed (although it’s under attack these days for different reasons), it seems as though Tesla has gotten away with a deplorable safety record because the details of its crashes have gone unlearned or never mattered after all. It’s reasonable to guess that Xiaomi — moving fast in Elon Musk mode— might be replicating mistakes already made by Tesla.
SU7 under the hood
Xiaomi’s first vehicle, the Xiaomi SU7, was announced in December 2023 and officially released the following March in Beijing.
Despite aspirations among Chinese OEMs for a vertically integrated business model, Xiaomi’s first vehicle uses many parts and components from abroad.
Its ADAS, branded as Xiaomi Pilot, uses a vision-based sensor array that deploys a single Nvidia Drive Orin X SoC with 254TOPS of computing power.
The SU7 that crashed was the “standard” SU7 model. High-end SU7 models use two Orin SoCs. The vehicles are upgradeable via OTA to the NOA feature in “urban conditions.”
It’s not clear whether the ADAS software stack in Xiaomi’s SU7s is domestic or imported. Yue Ma, author of a new book, “The Xiaomi Formula,” believes Xiaomi is using a stack by Deepmotion, which Xiaomi acquired in summer 2021. Reportedy, Deepmotion is a self-driving car startup founded in 2017 by four computer scientists from Microsoft Research Asia.
Crash details we don’t know yet
Pierrick Boulay, senior market and technology analyst at Yole Group, posed several big questions for which there is yet no answer:
Was the fire caused by the battery? If so, what contributed to its ignition?
Were the emergency door handles accessible and functional, or did they fail under crash conditions?
Boulay noted, “There is also an over-reliance on L2++ systems like what we saw in the past with Tesla FSD. These systems remain assistance systems and are not defined to replace the driver.”
Yu Yang, principal analyst at Yole Group, shared multiple concerns.
The accident happened while roadwork was being done on a highway.
He explained, “Normally there should be consecutive speed signs to lower the speed limit from 120kph to 50kph. But the car kept in Navigate on Autopilot, driving at close to 120kph, until it was rather close to the actual roadwork. The speed limit signs were neglected or not placed correctly. No evidence yet.”
What was the status of the driver before the accident?
“SU7 is equipped with a Drive Monitoring System (DMS). But [there was] no information on [whether] DMS gave warnings … before the accident to say the driver was careful or distracted (or too tired),” Yu suggested.
Automatic Emergency Braking (AEB) was not functioning.
Yu said, “According to Xiaomi, the AEB is valid between 8-135kph (which is a pretty wide range). But what it can detect is limited [to] vehicles, pedestrians, and two- and three-wheelers.”
“So,” he concluded, “it can’t prevent a crash like this.”
Based on Xiaomi’s announcement, the car’s NOA shut down, alerting the driver to take over, only 3 seconds before the crash.
Yu observed, “The driver braked down to ~100kph (from close to 120kph) right before the deadly crash. This is clear proof the L2 system is not capable of handling a roadwork situation.” Yu added, ominously, “The autopilot could not even tell that it is incapable [of doing the job].”
What is the main reason for this failure?
Yu suspects that key sensors are missing. Among the issues, he cites the low computational power of the SoC and defects in the ADAS algorithm. He reiterated that drivers must be informed that L2 is not autonomous. It is assisted driving.
Yu also shared concerns about the functional safety of flush door handles, the safety level of EV batteries, and others.
Bottom line:
Speculation is already rife in China that safety-conscious Chinese consumers will soon start demanding more sensor technologies – such lidars by China’s Hesai or RoboSense Technology.
But the larger, lingering dilemma is whether Xiaomi will ever choose to admit the limitations of today’s NOA SU7 basic model and how this candor—or absence of it—will affect consumer attitudes on the safety of highly automated vehicles.
Yole’s Yu noted that China is in final preparation of L3 legislation this year. He said the accident might trigger some second thoughts. “However,” he concluded, “I think the trend would be slowed a little bit but not halted.”