When roads think and rails learn: AI makes transport smarter, faster, and greener

David Xu, Vice President of Huawei's Smart Transportation Business Unit

As global transportation systems evolve and expand, their focus is no longer just on building more, but on building better. Safer, smarter, more efficient, and sustainable transport networks are the name of the game. At the core of this change is digital transformation, powered by AI.

Whether it’s China’s push toward innovation-led logistics, the EU’s Smart Mobility Strategy, or Singapore’s next-gen land transport plans, it’s clear that data and intelligence will reshape the future of transportation.

Unlocking new transport productivity

Traditional transportation systems were designed around physical assets such as rails, roads, and runways. But the next generation of that hardware will run on data.

Even today, diverse modes of transport, from rail to low-altitude drone logistics, already generate terabytes of structured and unstructured data each day. The IoT, inspection video, GPS and telemetry coordinates, and customs documentation all populate these data streams.

Processing and acting on that information in real time is the key to unlocking what Huawei calls “new forms of transportation productivity.” It starts with a layered architecture built on sensing, cognition, and decision-making, all of it enabled by cloud-edge-device collaboration and AI large foundation models.

Here’s a look at how this sophisticated technology is being deployed in real-world systems.

Sensing: turning infrastructure into intelligent agents

Computer vision (CV) and IoT have turned static transport assets into active data generators. In China’s rail freight system, CV models trained on tens of millions of images identify 430 types of mechanical errors across over 60 train models, with sub-millimeter precision and 99.3% inspectional accuracy. Enabled by Huawei’s Ascend servers and 3B-parameter models, the system reduces inspection time from several hours to less than five minutes.

At Tianjin Port, on China’s northeastern coast, AI large CV models are trained first on broad visual data, then refined for port operations using a billion real-world samples. Today, they automatically detect safety threats like loitering or wrong-way driving at altitude, thereby ensuring safe, autonomous port operations.

This AI-first approach addresses industry-scale challenges such as aging infrastructure and workforce shortages, while increasing resilience and operational throughput.

Cognition: from document parsing to real-time traffic flow

Large language and vision models (LLMs and multi-modal encoders) are being used to tackle one of logistics’ oldest pain points: documentation. Customs declarations are often filled out in multiple languages and plagued by poor handwriting, ink smudges, different writing systems, and inconsistent formats. As a result, they have historically required 10–15 minutes of human scrutiny per form.

Huawei’s logistics-specific models, trained on tens of millions of documents and fine-tuned via prompt engineering and multi-modal pretraining, now achieve 96% recognition accuracy on known formats and 90% on previously unseen layouts.

In cities, cognitive AI is transforming traffic management. One city-level system with over 600 small-model algorithms is capable of orchestrating tens of subsystem interactions under a single AI agent umbrella.

Using real-time scene analysis, AI Agent leads each task—whether detecting jaywalkers or optimizing the timing of traffic lights—to the most effective algorithm. This bridges older systems and new AI without costly overhauls in the overall technology stack. Intent can be recognized and algorithm matching can be evaluated for scenario-specific problems. With multiple objects identified in a single image, each subsystem can be invoked to enable collaboration among algorithms and linkage between system applications, thereby ensuring road traffic safety and protecting customers' investments.

Solving for complexity in real time

Transport operations involve problems related to systems with lots of variables: flight and train scheduling, metro headway adjustments, port crane assignments, crew rotations, and so on.

Traditionally, these problems were solved manually—often over a period of days—resulting in delays and suboptimal utilization. Today, systems like Huawei’s OptVerse AI Solver combine multi-objective optimization with scenario-specific constraints to compute globally optimal solutions in seconds.

In one metro system in Eastern China, Huawei has systematically categorized and digitized hundreds of operational dispatch scenarios that used to take 15 minutes to resolve. Now, those issues can be resolved in no more than 15 seconds, even with millions of variables and rules in play. This enables real-time responsiveness to disruptions (e.g., signal faults, equipment failures) during peak demand, significantly improving reliability and smoothing out passenger flow.

Similar models are being deployed in EMU maintenance scheduling, container yard sequencing, and airport slot allocation—where optimization at scale creates measurable gains in throughput and service quality.

Building the compute backbone for intelligent transport

All of this requires serious backend infrastructure. AI in transportation is compute-intensive and latency-sensitive, demanding hybrid architectures that support edge inference, high-volume data ingestion, and centralized model training.

Huawei has addressed this with a system-engineering approach that unifies computing, networking, and storage resources—breaking through previous bottlenecks in model training and real-time inference. With DataArts and ModelArts as huawei specific AI model tool chain, model development cycles have shrunk from months to weeks. Meanwhile, compute, transmission, and storage layers are now co-designed to handle terabyte-scale daily data flows from hubs, fleets, and networks.

From experience-driven to data-driven

The transportation industry has long relied on institutional knowledge and manual workflows. But the combination of AI, cloud-native infrastructure, and real-world data has kicked off a new era of autonomous decision support at scale.

By integrating perception, cognition, and control into one intelligent loop, transportation systems can now adapt in real time—boosting safety, reducing emissions, and enhancing user experience.

In short, transportation is evolving into a self-optimizing system. Moreover, AI acts not just as a supporting tool, but as a core control technology that helps transport operators thrive in the next generation of mobility infrastructure.