As combined assisted driving gradually becomes a reality, the path to autonomous driving has become increasingly clear, with individual vehicle intelligence emerging as the mainstream direction. Individual vehicle intelligence involves equipping vehicles with an array of LiDAR sensors, millimeter-wave radars, high-definition cameras, and high-performance computing chips, enabling them to independently perceive their environment, make decisions on routes, and execute maneuvers.

However, individual vehicle intelligence frequently encounters physical limitations in practical applications. Visual sensors cannot penetrate large vehicles ahead, LiDAR's detection range is significantly reduced in heavy rain, and there are perception gaps when facing blind spots at intersections.

To address these shortcomings, the concept of vehicle swarm intelligence has emerged. It envisions creating a shared network for traffic environments with broader coverage and deeper perception dimensions through real-time information exchange among vehicles. Its logic is similar to real-time road sign warnings but with significantly higher precision and response speed. Achieving vehicle swarm intelligence is not easy, as many issues need to be resolved.

Technical Challenges Need to Be Overcome

The goal of vehicle swarm intelligence is to break the perception boundaries of individual vehicles, expanding the traffic environment from 'what I see' to 'what everyone sees.' Through vehicle-to-vehicle information exchange technology, vehicles can obtain real-time road conditions or pedestrian movements in blind spots kilometers ahead, enabling beyond-line-of-sight perception.

For high-speed autonomous vehicles, the decision-making system has an extremely low tolerance for latency, ideally below ten milliseconds. In existing cellular network environments, data forwarding through base stations introduces significant delays. Although 5G technology accelerates communication efficiency, ensuring that hundreds or thousands of data packets are accurately exchanged within milliseconds without conflicts in high-density traffic remains a significant challenge for communication architectures.

Beyond transmission speed, the depth of data fusion also determines the practicality of vehicle swarm intelligence.

Current collaborative perception solutions mainly fall into three categories: object-level, feature-level, and raw data-level. Object-level fusion only transmits structured information after recognition, such as a vehicle's coordinates and speed. This method has low bandwidth pressure but can lead to information loss or misjudgments due to varying algorithm accuracies among different automakers.

Achieving higher-order raw data fusion, where vehicles share radar point clouds or high-definition video streams, would generate massive data throughput. Current in-vehicle communication hardware and wireless spectrum resources cannot support real-time sharing on this scale.

Moreover, the reliability of perceived information is also a challenge. If a vehicle in the swarm transmits incorrect road condition information due to sensor contamination or malfunction, how should other vehicles relying on this information filter and verify it through their safety mechanisms? The industry has yet to establish unified fault-tolerance standards.

When vehicles attempt to transition from simple perception sharing to collaborative decision-making, the complexity of the problems grows exponentially. An individual vehicle only needs to consider its optimal path when making decisions, whereas in a vehicle swarm intelligence environment, the system must calculate a global optimal solution.

This involves collaboration among intelligent driving systems of different brands and levels. When a vehicle equipped with a high-level intelligent driving system encounters a vehicle with only basic assisted driving capabilities on a narrow road, how can they reach a consensus on who proceeds first through communication? Currently, manufacturers' planning and control algorithms are like black boxes, lacking a unified 'traffic language.' As a result, even if vehicles establish connections, they cannot understand each other's driving intentions, let alone collaborate.

This communication gap caused by inconsistent underlying technical logic keeps vehicle swarm intelligence stuck in the basic (junior) stage of 'mutual reminders' and unable to advance to the advanced (senior) stage of 'joint actions.'

Commercial Interests Are Hard to Overcome

While technical challenges can be addressed through research and development, commercial barriers among automakers pose the greatest obstacle.

In the current competitive landscape of intelligent driving, data is regarded as automakers' core asset and competitive moat. Leading manufacturers have accumulated vast amounts of extreme-scenario data through millions of kilometers of real-world testing to train their perception models and end-to-end algorithms.

Under the framework of vehicle swarm intelligence, this real-time perception data, originally considered private property, would need to be shared on a public platform. For automakers, this not only means the leakage of core technologies but could also erase their unique advantages in intelligent driving performance.

Each manufacturer tends to promote its system, attempting to make its intelligent driving solution the de facto industry standard. This market competition-driven 'stovepipe' data silos make cross-brand data interoperability extremely difficult.

In the early years, the autonomous driving sector experienced investment frenzy, with companies having sufficient funds to explore long-term technological paths. However, as capital becomes more rational and investment amounts continue to decline, automakers' focus has shifted from 'future visions' to 'immediate survival.'

To boost sales in the short term, manufacturers are racing to popularize urban NOA (Navigate on Autopilot) functions, most of which are based on optimizations of individual vehicle intelligence. This is because individual vehicle solutions have shorter commercialization cycles and lower reliance on external infrastructure.

In comparison, while vehicle-road-cloud integrated collaborative solutions have higher potential, their return on investment is difficult to demonstrate in the short term due to the need for coordination among governments, operators, equipment providers, and multiple automakers. This lack of immediate benefits reduces automakers' motivation to participate in collaborative construction.

Another manifestation of this misalignment of interests is the deadlock between 'vehicles waiting for roads' and 'roads waiting for vehicles.'

Automakers argue that if road infrastructure is incomplete or inconsistent, adding communication modules to vehicles will only increase costs without improving user experience. Meanwhile, road construction authorities believe that if the penetration rate of intelligent connected vehicles on the roads is insufficient, the intelligent roadside equipment built with huge investments will remain idle.

Currently, the penetration rate of connected vehicles in key demonstration zones is still low, and a large-scale commercial closed loop (closed loop) has yet to form. This has led to a severe disconnect between value creation and value acquisition. Without a clear profit model, even financially robust automakers lack the incentive to break down brand barriers and achieve true data openness and vehicle swarm collaboration.

Stringent Legal and Data Governance Constraints

Beyond technology and costs, data security and privacy protection are red lines that vehicle swarm intelligence must not cross. The essence of vehicle swarm intelligence is the flow and aggregation of data, much of which contains sensitive information. Vehicles' real-time movement trajectories are explicitly defined as sensitive personal information under Chinese law, and their processing is strictly regulated by the Personal Information Protection Law of the People's Republic of China and the Provisions on the Administration of Automobile Data Security.

When vehicles upload location data to the cloud or share it with surrounding vehicles in real-time for collaboration, ensuring that this data is not used for profiling or tracking, obtaining separate user consent, and providing convenient withdrawal options without compromising driving safety are all issues that need to be considered.

During vehicle swarm collaboration, vehicles also collect large amounts of video and image data from outside the vehicle. This data sometimes includes pedestrians' facial information and surrounding vehicles' license plate information. According to compliance requirements, such information must be anonymized before being transmitted outside the vehicle, such as by blurring faces and license plates through desensitization techniques.

This processing not only imposes additional requirements on in-vehicle computing power but may also affect the accuracy of collaborative perception due to the degradation of image quality caused by the desensitization process. Moreover, when data flows among different entities (automakers, roadside operators, cloud control platforms), defining responsibility chains becomes extremely difficult in the event of data leaks or hacking attacks.

National security considerations also impose boundaries on data flow. Geographic information and vehicle traffic data related to important sensitive areas such as military management zones and party-government organs are classified as important data and prohibited from being freely transmitted or taken out of the country.

For multinational automakers with global data centers, this means establishing completely isolated collaborative networks within China and undergoing strict security evaluations by relevant authorities. While this stringent management of data flow safeguards national security and personal privacy, it objectively increases the construction costs and operational difficulties of vehicle swarm intelligence systems. Until a technical solution is found that enables efficient data sharing while perfectly avoiding legal risks, all parties remain cautious in promoting vehicle swarm intelligence.

Final Thoughts

The development of vehicle swarm intelligence is not an overnight process. While individual vehicle intelligence has its limitations, it excels in independence and closure. Vehicle swarm intelligence, although logically flawless, is constrained by the complexities of the real world. To achieve true vehicle swarm collaboration, it may be necessary to start by establishing unified standards for cloud control infrastructure platforms, supplemented by comprehensive regulations on data ownership and sharing. Only when the four pillars of technology, commerce, economics, and law are firmly in place can vehicle swarm intelligence become a reality.

-- END --