Beijing | Industry Insight
SHINDEV Research Institute has recently observed a rapid rise in public discussion surrounding Apollo Go, Baidu’s autonomous ride-hailing platform. Alongside this growing attention, the underlying L4 autonomous driving technology has also become a focal point of industry-wide debate. Long regarded as a high-investment, hard-to-commercialize frontier technology, L4 autonomous driving is now presenting a tangible, real-world case that invites renewed reflection on its business models and scalable deployment paths.
Within the SAE automation framework, Level 4 autonomy enables vehicles to operate fully under system control within designated conditions, without requiring human intervention. However, this very characteristic of “driverless operation” has historically made L4 one of the most difficult and costly technologies for large technology companies to sustain. Complex scenario design, high system costs, regulatory uncertainty, and unclear liability allocation have previously forced several leading players to scale back or abandon L4 development altogether.
Against this backdrop, Apollo Go has emerged as a landmark case by achieving large-scale, real-world deployment of L4 autonomous driving through the ride-hailing scenario.
According to SHINDEV Research Institute, before the rise of Apollo Go, the autonomous driving industry had witnessed multiple high-profile exits from the L4 track, triggering skepticism over whether L4 autonomy was commercially viable at all.
From a technology readiness perspective, Level 3 and below (including L2 and L2+) autonomous driving systems have reached relative maturity in terms of safety and application, and are now widely deployed across new energy vehicle models. Due to regulatory constraints, however, automakers generally market these capabilities as L2+. In contrast, although L4 represents only a single-step advancement in classification, it entails exponential increases in system complexity, redundancy requirements, and compliance challenges.
Duanzhiqiang, President of SHINDEV Research Institute, notes that the prolonged difficulty in commercializing L4 is not due to technical infeasibility, but rather a lack of suitable application scenarios.
On the one hand, L4 systems require extensive road testing, high-definition mapping, and continuous data feedback for model training. On the other hand, the complexity of urban road networks and current regulatory frameworks limit L4 operations to specific, localized environments.
More importantly, high-level autonomous driving typically depends on multiple LiDAR units and high-performance computing platforms, driving system costs far beyond those of mainstream intelligent driving solutions. For most consumers, high-priced vehicles with restricted usability fail to generate sufficient demand—making the C-end passenger vehicle market an unsuitable entry point for L4.
Duanzhiqiang emphasizes that while L4 struggles in private passenger vehicles, it naturally fits B2B scenarios such as ride-hailing.
Ride-hailing vehicles operate on relatively fixed routes, making HD map development and repeated system training more feasible. Moreover, removing human drivers significantly reduces labor costs, partially offsetting high technology expenses. Once fleet size reaches scale, operational breakeven—and even profitability—becomes achievable.
It is precisely under this logic that Baidu has successfully demonstrated a viable L4 business model through Apollo Go, providing the industry with a verifiable reference path.
SHINDEV Research Institute believes the commercial adoption of L4 autonomous driving can be evaluated through three core drivers:
rising labor costs driving automation demand;
high-safety-requirement scenarios with elevated accident costs;
and closed or semi-closed environments where efficiency gains are mission-critical.
Beyond ride-hailing, public transportation, trunk logistics, sanitation services, municipal delivery, ports, and mining operations are all promising candidates for early L4 commercialization.
Industry participants can generally be categorized into three groups:
automakers and system integrators such as Huawei, BYD, and XPeng;
chip and computing platform providers such as Horizon Robotics;
and third-party autonomous driving technology and operation platforms led by Baidu.
Duanzhiqiang notes that technical capabilities across these players are relatively comparable. The decisive differentiator lies in access to viable scenarios and sustainable business models. “Baidu’s significance is not purely technological,” he states. “It lies in proving that L4 autonomy can generate revenue—an outcome that will inevitably attract more entrants.”
Although Apollo Go has achieved scaled deployment, early user feedback in certain cities reveals conservative perception and cautious decision-making behaviors—an issue rooted in current technological trade-offs rather than company-specific shortcomings.
Today’s autonomous driving solutions generally follow two main approaches:
pure vision-based systems represented by Tesla’s FSD, which are cost-efficient but sensitive to environmental conditions;
and multi-sensor fusion systems leveraging LiDAR and HD maps, as used by Apollo Go, which offer strong perception but at higher cost and with conservative intelligence strategies.
Duanzhiqiang emphasizes that L4 autonomous driving is fundamentally a highly collaborative systems engineering challenge, requiring integrated development across four dimensions:
the vehicle as the core platform,
energy as the foundation,
road infrastructure as the carrier,
and cloud computing as the intelligence backbone.
Recent national initiatives accelerating vehicle–road–cloud integration, including the announcement of pilot cities by China’s Ministry of Industry and Information Technology, have laid crucial groundwork for broader L4 deployment.
SHINDEV Research Institute concludes that L4 autonomous driving is not a short-term disruptive leap, but rather a long-cycle industrial evolution requiring sustained investment and incremental progress. The true value of Apollo Go lies in validating a sustainable commercial framework.
Ultimately, success will belong to those who can establish closed-loop operations across diverse scenarios. The widespread adoption of L4 autonomy will depend not on isolated technological breakthroughs, but on the systemic maturity of scenario selection, cost structures, policy coordination, and ecosystem development.
