Where PHEV Meets AI: Zeekr 9X's Experiment with Physical World Models

When a car starts understanding the physical world, is it still just a mode of transport?

The Zeekr 9X five-seater version has arrived at dealerships, launching on July 28. On the surface, it looks like the routine rhythm of an electric-hybrid SUV, but combined with Zeekr's technical roadmap and the current trend of fusing AI with the physical world, this car deserves to be dissected from an information theory perspective.

The electric-hybrid architecture itself is an information processing system. The energy flow scheduling between the engine, motor, and battery is essentially an optimal coupling problem of multimodal information. Zeekr 9X opening the new "Exploration Brown" color for both five- and six-seater versions simultaneously seems like an aesthetic option, but it actually reflects the automaker's refined modeling of user scenarios—different seat layouts correspond to different load distributions, directly impacting energy management strategies. Traditional automakers relied on calibration tables; the AI era relies on real-time online learning.

More critical is intelligent driving. Zeekr's NZP (Highway Navigation Pilot) has accumulated vast road scene data, and as a flagship SUV, the 9X likely carries higher-end perception hardware (LiDAR, millimeter-wave radar, camera arrays). From the perspective of world models, these sensors collect high-dimensional physical state vectors: position, velocity, acceleration, friction coefficient, pedestrian intent. The vehicle needs to infer causal structures in real-time from these vectors—for example, whether the car ahead braking hard is due to slippery roads or obstacles ahead. This is precisely the cutting-edge topic of Physical AI: extracting latent variables of dynamics from visual streams.

In-cabin AI interaction is also noteworthy. Zeekr's AI Eva is evolving toward embodied intelligence, not just acting as a voice assistant but understanding the physical environment inside the car: passenger posture, temperature distribution, lighting conditions. The five-seater layout makes the rear space more flexible, suitable for family scenarios. The AI needs to dynamically adjust air conditioning ducts, seat angles, and even audio equalization to match the physiological states of different occupants. This isn't a simple rule engine, but an adaptive strategy based on reinforcement learning.

From an industrial trend perspective, the electric-hybrid route hasn't become obsolete in 2025; instead, it has gained new vitality through AI intervention. Range anxiety in pure EVs cannot be fundamentally solved physically, but AI can approach theoretical limits by predicting driving behavior and optimizing energy allocation. The launch of the Zeekr 9X essentially elevates AI from "an app on the infotainment system" to "the core layer of vehicle physical control."

Prediction: By 2026, high-end electric-hybrid SUVs will come standard with world model inference engines supporting online learning. The Zeekr 9X will be the first mass-produced sample of this trend. Not because it is particularly smart, but because problems in the physical world ultimately require Physical AI to solve.

Original Link: https://www.ithome.com/0/981/282.htm