Artificial Intelligence Learns to ‘Read’ Roads
Researchers at the Moscow Automobile and Road Construction State Technical University (MADI) have unveiled MADI-Speed-AI, a platform in which artificial intelligence handles nearly the entire engineering workflow – from analyzing documents to identifying hazardous road sections.
MADI’s Automated Control Systems Department developed MADI-Speed-AI, an AI-native platform for modeling vehicle speed profiles. Its purpose is to support road design, analyze traffic flows and identify potentially hazardous locations in advance. What sets the system apart is that AI does not simply offer suggestions but effectively orchestrates the entire workflow: an LLM agent converts a task described by an engineer in natural language into a formalized query; separate AI agents extract parameters from files and check them against GOST standards and logical constraints; and a deterministic mathematical core then performs the actual calculations of vehicle dynamics and speeds. An AI analyst completes the cycle by interpreting the results, identifying high-risk sections and producing preliminary conclusions.
The platform is built on a microservices architecture, with a database containing 42 vehicle models and 156 road profiles covering a total of about 1,200 kilometers (745 miles). Its calculations achieve 98.7% accuracy compared with real-world data. The developers stress, however, that the system remains a research prototype rather than a production-ready tool for making safety decisions.
The Platform’s Next Step
The main prospect for MADI-Speed-AI is its evolution into an industrial-grade engineering solution. The developers plan to enable the AI agent not only to identify problems but also to suggest design modifications that could improve safety.
Integrating the platform with computer-aided design (CAD), geographic information systems (GIS) and digital twins of highways could be particularly valuable. MADI-Speed-AI would then become part of a unified digital environment rather than an isolated calculation tool: an engineer could upload a project, while the system automatically analyzes road geometry, models the movement of different vehicle types and highlights problem areas before construction or reconstruction even begins.
The Road to Smart Design
Russia’s road sector has adopted digital technologies in stages. Beginning in 2021, intelligent transportation systems (ITS) were introduced in 22 Russian regions under a national project, with the goal of automating traffic management and improving safety. By 2024, such systems were operating in 62 urban areas, analyzing road congestion and identifying unsafe sections.
Another important step was the creation of a digital twin of the M-11 Neva highway in 2023–2024. At the same time, MADI and Ediny Operator (Unified Operator) studied how digital infrastructure could help prevent accidents on toll highways – an area closely aligned with the core purpose of MADI-Speed-AI.
Similar approaches are gaining momentum worldwide. Google Project Green Light, for example, uses AI and mapping data to model traffic at intersections and recommend changes to traffic-light timing. In Beijing, intelligent traffic signals automatically adjust the duration of green lights according to traffic flows, reducing the road congestion index by about 19%. These examples show that smart management of transportation and infrastructure has become a global trend, and the Russian platform fits into that shift while emphasizing engineering reliability and combining AI with proven computational methods.
A Place in the Design Ecosystem
MADI-Speed-AI reflects several important trends in Russia’s IT sector at once: the development of industry-specific AI agents, specialized engineering software and digital twins, as well as the shift from individual AI models toward integrated platforms. Its defining feature is a clear separation of roles: generative AI handles task interpretation, document processing and analysis of results, while critical calculations remain the responsibility of a deterministic mathematical model. This approach is better suited to safety-critical engineering systems, where generative models cannot be relied on entirely.
In the near term, the most realistic path is to validate the platform on actual road projects, build an evidence base and integrate it with Russian GIS, CAD and digital-twin systems. If testing proves successful, MADI-Speed-AI could become a competitive domestic product for road designers and operators and, eventually, part of a national digital design ecosystem for transportation infrastructure, where forecasting and safety are built into projects from the drawing-board stage.
