Jan. 05, 2026
Source:
At a critical stage in the scaling of autonomous driving, L4 autonomous delivery vehicles (Robovan) are rapidly transitioning from “technology validation” to “commercial operations.” As a leading player in this sector, Rino.ai continues to drive technological innovation, making breakthroughs in perception architecture, mapping dependency, hardware platforms, and future development paths, demonstrating both industry-leading technical capabilities and strategic vision.
Since the second half of 2024, leading players in the autonomous delivery industry have been deploying BEV (Bird’s Eye View) perception frameworks on their vehicles. This Tesla-inspired approach fuses multiple cameras to create a unified spatial perception, significantly enhancing the environmental understanding of pure vision systems.
Rino.ai not only successfully implemented the BEV framework but also achieved advanced engineering optimizations: lateral and rear auxiliary LiDARs were completely removed, leaving only a single forward-looking radar for safety redundancy.
Unlike competitors who rely heavily on mechanical LiDARs—expensive and less reliable—Rino.ai in 2025 became the first to switch its main LiDAR to vehicle-grade, fully eliminating non-vehicle-grade mechanical LiDARs. This milestone demonstrates that Rino.ai’s perception system has reached high stability, laying a solid foundation for large-scale mass production of autonomous delivery vehicles.
High-definition maps have long been considered standard for L4 autonomous driving, but their high collection and maintenance costs have become a barrier to large-scale deployment. The industry is now increasingly moving toward a “light map” approach, minimizing map elements and simplifying production processes during mapping.
Rino.ai has adopted a pragmatic path: optimize mapping first, then data collection. Through deep collaboration and joint development with Baidu Maps, the company has completed real-world vehicle testing in multiple locations and officially deployed the solution on its vehicles. This approach significantly reduces manpower and time costs for map creation, while ensuring the safety and generalization capabilities of L4 systems in complex urban scenarios.
Importantly, Rino.ai is not blindly pursuing a “fully map-free” system at this stage. Instead, it carefully manages map dependency based on the reality that L4 autonomous delivery vehicles operate without human drivers as a safety fallback. This technical restraint demonstrates the company’s commitment to safety and social responsibility. Rino.ai also believes that a fully map-free system is currently not achievable with existing architectures, and plans to achieve true map-free operation through end-to-end architectural upgrades.
In terms of computing power, the industry generally relies on 500T platforms to support BEV model operations. Rino.ai, however, took a more challenging path, successfully deploying a high-performance BEV perception model on a 200T domain controller.
This achievement reflects Rino.ai’s deep expertise in model compression, quantization, and deployment, and highlights its outstanding engineering capabilities in AI infrastructure. Lower computing power translates to reduced energy consumption, cost, and thermal stress, delivering higher cost-efficiency and reliability for autonomous delivery vehicles operating in high-frequency urban last-mile scenarios.
Rino.ai is the first company in the industry to complete vehicle-grade SOP (Start of Production) for L4 autonomous delivery vehicles. By partnering with OEMs such as Xinyuan Auto and Jingwei Hengrui, Rino.ai has developed truly vehicle-grade L4 autonomous delivery products and technology solutions. From sensors to domain controllers, and from electric drive systems to drive-by-wire chassis, all components meet vehicle-grade standards, not only satisfying functional safety requirements but also ensuring long-term stability under high-intensity operations.
Moreover, this vehicle-grade platform lays a solid foundation for a more efficient “data flywheel”. By standardizing hardware first, Rino.ai enables a closed-loop, data-driven model iteration cycle. As the fleet scales rapidly, high-quality, standardized data accelerates algorithmic updates, creating a hardware–data–model virtuous cycle that drives continuous improvement in autonomous delivery performance.
Although BEV + light map is the current industry standard, the consensus is that incremental optimization alone cannot achieve true map-free L4 autonomous driving. Drawing inspiration from breakthroughs in end-to-end large models for passenger vehicle urban NOA, Rino.ai launched end-to-end technology development in 2025.
The team has already completed the core technology groundwork and plans to begin prototype system on-vehicle testing in the coming months, accelerating mass deployment. When realized, Rino.ai will be the first in the autonomous delivery vehicle sector to implement a truly map-free L4 solution—operating without high-definition maps and relying solely on standard navigation maps and real-time perception, capable of handling complex urban scenarios.
A company’s technology roadmap reflects its strategic vision and technical strength. Rino.ai has not chased short-term trends, but adheres to the principles of “safety first, mass-producible, and continuously evolving,” achieving simultaneous breakthroughs across four key dimensions: BEV perception, vehicle-grade hardware, light map evolution, and end-to-end architecture.
As the autonomous delivery industry enters the deep waters of large-scale operations, Rino.ai’s technological advantages are gradually translating into commercial competitiveness. Over the next five years, L4 autonomous driving is expected to experience leapfrog development, with autonomous delivery vehicles integrating on a large scale into urban and rural logistics scenarios, reaching a fleet size of millions of units and showing a clear trend toward global markets. Rino.ai is confident it will remain a leading force driving this wave forward.
