Rovers Are Getting Faster
Delivery robots are becoming a more familiar sight in busy urban environments, and they are getting faster. Over five months, the average speed of Yandex's rovers increased by more than 40%, rising from 3.5 to 5 km/h.
The speed gains come from a neural-network planner that works alongside a conventional algorithm. Both systems propose possible routes, and an algorithm selects the best option, improving reliability. Yandex also uses a virtual simulator to test and improve the system's reliability. More than 6,000 virtual rovers train simultaneously in this environment: in one hour, they accumulate the equivalent of about 140 days of experience; in 40 hours, they gain roughly 15 years. The simulator also allows the company to test rare and high-risk scenarios without putting people in danger.
Outlook and Export Potential
Yandex now operates a fleet of more than 1,500 rovers across 13 cities. In the third quarter, the robots completed more than 750,000 deliveries, bringing the cumulative total to over 2 million. Since 2019, the cost of robotic delivery has fallen fivefold and is now approaching the cost of using human couriers. By the end of 2027, the company plans to deploy 20,000 fourth-generation rovers on the streets of Russian cities.
Higher travel speeds clearly increase fleet productivity, but the overall economics also depend on maintenance, charging, mapping and remote supervision. That is why Yandex is developing a hybrid delivery model: a rover transports an order to a customer's building, while a human courier brings it to the apartment door. This approach could work particularly well at large sites such as university campuses, business centers and hospitals. The government has also established an experimental legal framework for delivery robots in 35 regions, making their use legally permissible.
The export potential of the new product, however, remains uncertain. Yandex has international experience, but it has not yet announced confirmed plans to supply the updated AI-powered model abroad. Even so, the Russian technology could attract international interest as an example of how trainable AI can be applied in commercial robotics. To enter foreign markets, the company will need to demonstrate the project's economic viability and show that the system can be adapted to different infrastructure.
A Look Back: From Pilot Programs to a Mass-Market Service
In 2021, Yandex rovers delivered food on a university campus in the United States. The following year, Russian Post tested the robots in Murino and Innopolis. By 2023, the joint pilot had ended: 36 robots had delivered more than 18,000 items, with 99% of deliveries completed on time. However, the project did not develop into a permanent service.
In 2024, market participants highlighted the technology's high cost and infrastructure constraints. A turning point came in 2025, when Yandex unveiled its fourth-generation rovers and moved into mass production.
In May 2026, the company introduced a neural-network planner that reduced delivery times by an average of 10%. The global market was expanding at the same time. By May 2026, Starship Technologies had completed more than 10 million deliveries across eight countries, demonstrating strong demand for the technology.
Forecast: The Race for Smarter Robots
Faster rovers mark another step in the shift from experimentation to large-scale deployment. Over the next few years, wider geographic coverage and higher delivery volumes will depend on mass production, advances in AI and the development of an appropriate regulatory framework. In the longer term, autonomous navigation algorithms could become a common foundation for different types of robots, while simulation-based training could become standard practice for safe autonomous systems.
Costly equipment, the need for accurate digital maps, urban conditions, weather and the continued need for remote supervision remain significant constraints. Even so, the next phase of competition in autonomous robotics is likely to be determined less by mechanical engineering than by AI performance: how well systems can anticipate changes in their surroundings and manage large fleets of machines efficiently.
