The autonomous driving industry is entering a new phase centered on "physical AI", as advances in world models and reinforcement learning begin to reshape how machines understand real-world environments, said Yu Qian, co-founder and CEO of QCraft.

"Autonomous driving is the best entry point to physical AI," said Yu at the 2026 Intelligent Electric Vehicle Development Forum in Beijing, arguing the industry’s focus is shifting from building "systems that drive" to creating a "brain for the physical world".

Yu said vision-language-action models, combined with world models and reinforcement learning, will form the core technology stack from 2026, enabling systems to better grasp physical rules, social norms and cross-scenario reasoning. QCraft plans to disclose further progress at the Beijing auto show later this year.

The company is commercializing the approach through its QPilot 2.0 driver-assistance system, which uses a unified architecture to support highway and urban navigation features across different price tiers.

Yu said the strategy emphasizes single-chip optimization and tighter hardware-software integration, in contrast to industry reliance on higher computing power.

QCraft is also deploying an L4 Robovan for unmanned logistics, targeting applications from parcel delivery to cold-chain transport.

The vehicle can operate around the clock and reduce route-level operating costs by about 50 percent, according to the company. The system has already been rolled out in several Chinese cities.

Yu said QCraft is pursuing a dual-track strategy across L2 and L4 systems on a shared technology base, adding that 2026 could mark the start of a "golden decade" for autonomous driving.