The secret behind Radish Run's "stall": the autonomous driving network got "stuck"

If you’re trading stocks, look at Jinqilin analysts’ research reports—authoritative, professional, timely, and comprehensive—to help you uncover high-potential theme opportunities!

Source: Caixin News

Zhu Xichan pointed out that the communications network supporting Robotaxi operations is not the legendary 5G communications; it is still the consumer-grade network that we use on our phones. Lagging and dropouts cannot be avoided. Safety issues need to be addressed, and technological progress should be met with tolerance.

This article’s author: Zhang Xiaodi, She Yueming, Yao Fei

On the evening of March 31, 2026 at 8:57, on the Third Ring Road in Wuhan, a Robotaxi—an autonomous vehicle from Baidu’s Luobu Kuai Pao—suddenly slowed down. It then came to a complete stop in the middle of the express lane. Ten minutes later, not only this one car, but nearly a hundred white autonomous vehicles simultaneously went “dazed” and stopped on the main roads—on the Taizihu Bridge, the Baishazhou Bridge, the Second Ring Road, and the Yangsipu Bridge.

One passenger said they were trapped on the elevated road for nearly two hours. The car’s SOS button couldn’t be reached, and the customer service line was busy. In the end, police officers had to rescue them by walking up to the elevated road, going vehicle by vehicle.

The next day, Wuhan traffic police issued a report stating that they initially believed it was caused by a system malfunction.

In Wuhan, locals have given the Luobu Kuai Pao autonomous vehicles a nickname—“Shao Luobu.” In Wuhan dialect, “Shao” means “stupid.” This was not the first time Luobu Kuai Pao had “broken down.” In July 2024, during Wuhan’s evening rush hour, vehicles also suddenly stopped on the road. Ultimately, after traffic police called the service line, the passengers sat in the driver’s seat and drove the car to the side of the road. The same “system malfunction,” the same “traffic police as the fallback,” nearly two years later—yet again.

Luobu Kuai Pao is Baidu’s core business in intelligent driving and mobility services, and an important direction for Baidu to commercialize AI technology. According to Baidu’s 2025 annual report, the operating entity of Luobu Kuai Pao is Luobu Yunliang (Beijing) Technology Co., Ltd. This company is wholly owned 100% by Apollo Intelligent Technology (Beijing) Co., Ltd., a subsidiary of Baidu (abbreviated as “Apollo”).

As of now, neither Baidu nor Luobu Yunliang has publicly explained the cause of this malfunction. Caixin News contacted Baidu, and the company said it has no relevant information available to disclose.

In a research report released by Guidehouse Insights, an industry analysis organization for the autonomous driving sector, in the fourth quarter of 2025, Baidu Apollo was listed as one of two global leaders in the autonomous driving field, with the other being the U.S. company Waymo.

According to the introduction, Luobu Kuai Pao’s autonomous driving technology is supported by Baidu’s four-layer, full-stack AI architecture (cloud infrastructure, deep learning frameworks, large models, and applications).

As of February 2026, its global footprint has expanded to 26 cities. Since February 2025, in all operational cities in mainland China—including Beijing, Shanghai, Shenzhen, Wuhan, Chengdu, Chongqing, Haikou, Sanya, and others—it has achieved 100% fully driverless operation (i.e., no safety operator inside the vehicle), equivalent to Level 4 (L4). It has also obtained paid operation licenses in multiple cities. To date, the number of autonomous driving mobility orders provided to the public has exceeded 20 million.

On the international market side, it has also entered cities such as Dubai, Abu Dhabi, London, Saint Gallen, and Seoul. On March 30, Luobu Kuai Pao officially launched commercial driverless operations in Dubai. But the very next day in Wuhan, a scene of collective “breakdowns” appeared.

What caused this malfunction, and can the problem be resolved? Caixin News interviewed Zhu Xichan, a professor at Tongji University’s School of Automotive and Energy, and the chief expert at the China Automotive Engineering Research Institute for intelligent connected vehicles. He has long focused on automotive safety and intelligentization research, and is one of the main founders of the country’s first automotive crash safety standards system. In recent years, he has focused on intelligent driving testing and evaluation and standard-setting. As the head of i-VISTA China’s intelligent vehicle index evaluation system, he has deep insights into technology routes, safety bottom lines, and standards systems for intelligent driving.

Zhu Xichan has repeatedly voiced calm perspectives amid industry hype. He once stated directly, “If someone announces mass production and deployment for end-to-end from the start, then you can’t buy this vehicle,” and he also advocated a fusion architecture where “AI raises the ceiling, and rules backstop the bottom line.” He also criticized the “10-second takeover” requirement for L3 autonomous driving, arguing that safety must be addressed all at once.

The following is the interview:

Caixin News: The Wuhan traffic police initially judged that the cause of this incident was a “system malfunction.” From your professional perspective, what technical issues does this kind of situation expose in autonomous driving?

Zhu Xichan: Many people think autonomous driving is “the car can drive itself,” like human drivers—one driver, one vehicle, roaming everywhere. This is called “single-vehicle intelligence.” The currently operating Robotaxis worldwide are not on this path.

They follow the “connected intelligence” route. Today, the driver position on Robotaxi vehicles no longer has a safety operator, but in the cloud-based back-end supervision platform there are safety supervision staff. I went to visit the cloud supervision platform at the intelligent connected vehicle demonstration zone; the supervisors sit in a device similar to a driving simulator. On a large screen in front, they display the operation of the driverless vehicles. If problems occur, supervisors can intervene remotely to solve issues that cannot be resolved by the vehicle-side autonomous driving software.

Currently, the Robotaxi belongs to the category of connected intelligence. Put simply, five things work together in coordination: vehicle, road, cloud, network, and map.

Vehicle: The most important is the intelligent vehicle. It uses multi-environment perception sensors composed of cameras + millimeter-wave radar + lidar. It has AI compute support built on high-compute chips. It also uses a driving system built with a visual-language large model (VLA) + an end-to-end deep learning model.

Road: Intelligent roads collect road traffic information. On the map there is real-time road conditions, used for path decision-making of driverless vehicles.

Cloud: As mentioned earlier, there are back-end platform personnel supervising. In domestic demonstration operations, typically one supervisor can supervise 5 to 10 driverless vehicles. In the United States, Waymo also has back-end supervisors, but I haven’t seen the parameter for how many vehicles each person supervises.

Network: It refers to the communications network between the vehicle and the cloud.

Map: It refers to the navigation map, high-precision maps, and integrated localization resources—fusion positioning using satellite positioning, ground base-station positioning, and the vehicle’s inertial navigation, among others.

These five are all indispensable. If any link has a problem, the system will trigger a safety mechanism—stopping. On the evening of March 31, the issue is likely in the “network.”

Caixin News: So the problem isn’t in the “vehicle,” but in the “communications”?

Zhu Xichan: Yes. There is a so-called “public secret” in the industry here: at the time, connected intelligence technology proposed building a 5G communications network—using 5G base station construction to realize a high-reliability, super-bandwidth, low-latency real-time communications network. But the early construction of intelligent connected vehicle demonstration zones did not go well. The communications network used to operate Robotaxi is not the legendary 5G communications; it is still the consumer-grade network that we use on our phones. Lagging and dropouts still can’t be avoided.

The malfunction on the evening of March 31 was very likely caused by the communications network. The connection between the vehicle and the cloud was severed, the system triggered a safety mechanism, and vehicles within the network-failure area collectively stopped. This is not that the “vehicle is broken,” but that the “network is down.”

Caixin News: Then the vehicles stopped directly in the middle of the travel lane, rather than pulling over to the side—was that a design flaw?

Zhu Xichan: First, I want to confirm one point: Luobu Kuai Pao has safety fallback measures. When communications are cut off, the vehicle stops; it does not continue to “run naked” on the road. In terms of safety logic, that is correct. But we are not satisfied with the control logic of the safety measures.

The vehicle-side software and hardware have no fault. It has high-precision maps, knows it is on an elevated road, and also knows there are lanes next to it, exits, and roadside areas. As an L4-level intelligent driving system, it is capable of autonomous lane changing.

So what is the reasonable approach? After triggering the safety mechanism, it should drive the vehicle off the urban expressway and park at the roadside, remind passengers to get out, and then contact the supervision platform after reaching the safe area.

But now it stopped directly in the fast lane. That’s why you can’t blame people in Wuhan for giving it the nickname “Shao Luobu”—in Wuhan slang, “Shao” also means “stupid,” and it is well-deserved.

Of course, we should also have some tolerance for the development of artificial intelligence technology. When issues are found in Robotaxi demonstration operations, technology companies need to iterate quickly so that the technology can mature quickly and benefit society.

Caixin News: Many passengers have reported that the SOS button wouldn’t connect, customer service lines were busy, and ground staff didn’t arrive for a long time. What is the reason?

Zhu Xichan: If the communications network is normal, cloud-based supervisors can handle the situation quickly through remote control, which would have little impact on road traffic. But when communications are cut off, the back-end cannot be handled remotely. If the back-end cloud supervision platform cannot handle it, it can only rely on ground personnel—i.e., the “ground staff”—to go to the scene. That night, some “broken down” vehicles were on elevated roads, which by itself would cause congestion. Ground staff being stuck on the road makes it difficult to arrive in time; generally, it takes a long time.

Based on these pieces of information, Luobu Kuai Pao’s operations in Wuhan appear to comply with the management requirements of the intelligent connected vehicle demonstration zone. This is not a problem unique to Luobu Kuai Pao. Four months ago, a citywide power outage happened in San Francisco in the United States; Waymo’s driverless vehicles also collectively stopped at intersections and turned on hazard lights.

Caixin News: Then for the legal responsibilities of this “breakdown” incident, who should be held accountable?

Zhu Xichan: When it comes to responsibility and legal gaps, these operating Luobu Kuai Pao driverless taxis are vehicles approved and recognized for demonstration operations by the Wuhan Intelligent Connected Vehicle Demonstration Zone. At present, the responsibilities and legal clauses for conventional vehicles and road traffic cannot all be applied, and judgment must be made based on the management regulations of the intelligent connected vehicle demonstration zone.

Caixin News: Baidu’s annual report shows Luobu Kuai Pao has been operating in 26 cities nationwide, with cumulative orders exceeding 20 million. What do you think about the relationship between this expansion speed and public safety?

Zhu Xichan: Demonstration operations through intelligent connected vehicle demonstration zones are intended to expose problems—fully surface them, and improve them in a timely manner—so that products can move toward maturity.

Autonomous driving vehicles are one of the most valuable application scenarios for artificial intelligence and play a major role in the digital economy. We cannot kill technology exploration and technology development because of one or two incidents. At the same time, the safety requirements for autonomous driving are actively advancing—test technologies, the formulation of evaluation standards, and accepted certification methods, etc. Industry participants should each do their own job: technology companies should iterate and improve products; research institutes should innovate evaluation methods for intelligent agents; and government主管 departments should explore innovative approaches to safety supervision for intelligent agents.

Safety issues must be faced; technological progress should be met with tolerance. We should not allow safety hazards from new technologies to go unchecked, nor should we shout for punishment and suppression when something happens, thereby killing the development of new technologies.

Caixin News: What about safety redundancy?

Zhu Xichan: Of course, we also cannot ignore that the pace of building safety redundancy for Luobu Kuai Pao seems to be lagging behind the pace of commercial expansion. But is this a problem of Baidu alone? Not entirely. However, there are still relatively few industry participants at the moment, so you can’t say this is a “common problem.”

Caixin News: What do you think Luobu Kuai Pao should do next?

Zhu Xichan: Based on this incident, I have three suggestions.

First, Luobu Kuai Pao’s company should improve its public relations strategy. When problems occur, it should promptly issue official statements to explain the technical nature of the incident. Only transparent information disclosure can prevent public opinion from expanding without limit.

Second, optimize the “minimum risk strategy.” Based on the current information, after Luobu Kuai Pao triggers the safety mechanism, its “minimum risk handling strategy” should still be optimized. It is reasonable to terminate normal operation of driverless vehicles in the network-failure area once a communications network fault is diagnosed. But relying on the intelligent vehicle to drive itself off the urban expressway, find a safe roadside stop, and guide passengers to get out—that more reasonable minimum risk handling strategy should be incorporated into autonomous driving algorithms.

Finally, continue to explore more reliable communications solutions. In “vehicle-road-cloud-network-map,” the “network” is currently the weakest link. Even though early construction of 5G communications foundations at the vehicle-regulation level for “connected intelligence” encountered obstacles, communications technology in the connected intelligence solution for driverless driving should continue to be explored and improved, so as to explore a more reliable communications方案 for vehicle-to-cloud interconnection in “vehicle-road-cloud-network-map.”

Caixin News: Final question—if your family asked you, “Can I take Luobu Kuai Pao now?” what would you say?

Zhu Xichan: My attitude is: you can try it for daily commuting, but you need to be mentally prepared—it’s not smart enough yet.

		Sina Statement: This message is republished from a media partner of Sina. Sina.com publishes this article for the purpose of sharing more information, and does not mean it agrees with the viewpoints or confirms the descriptions. The content of this article is for reference only and does not constitute investment advice. Investors act at their own risk.

Huge amounts of information, precise analysis—available in the Sina Finance App

责任编辑：杨红卜