The Immersive Cockpit Feels More Like a Product Package Than Just Buttons

Li Auto's OTA 9.1.1 bundles the "Sensory Cockpit" and multiple experience optimizations into a single version, covering high-spec models of the L9, L8, and L6. It turns the cockpit experience into a product package that can be sold, continuously updated, and operated long-term.

When I pitch AI solutions to clients, my biggest fear is them only asking about model parameters and chips. Car infotainment users don't buy that. They ask: Can my wife in the passenger seat get jolted less by sudden braking? Can the AC and audio auto-adjust for stability when my kids are in the second row? When I'm stuck in traffic after work, can the car stop hesitating like a novice driver?

Added two assisted driving styles, "Efficient" and "Comfortable," and optimized the driving experience in scenarios such as starting on green lights, accelerating during lane changes, and bypassing obstacles.

Descriptions like this sell better than large embodied intelligence models. Because it translates abstract capabilities into driving preferences. Users won't pay for a large model, but they will pay for one less takeover, one less jerk, and one less parking scrape. This capability can be packaged as a product; the key is linking voice, seats, lighting, AC, maps, parking, and assisted driving styles into several small workflows.

I've recently been using some APIs to build small demos, and my takeaway is not to be greedy with context. Car systems are the same—caching all vehicle data makes costs and privacy look ugly; keeping only high-frequency intents, like who gets in the car, where they're going, whether to refuel, and parking preferences, actually stabilizes the experience.

From Huawei Cloud's perspective, car system OTAs are very similar to building enterprise AI solutions. The front end is a button; the back end involves sensors, maps, edge-side models, cloud training, gray releases, rollbacks, and customer service scripts. Li Auto 9.1 has already released 31 new features and 10 experience optimizations, and 9.1.1 continues to add the Sensory Cockpit and optimizations—the pace isn't slow.

But fast pace means problems arise quickly too. The more features there are, the more likely users are to ask: How exactly do I use this? Which cars support it? Why hasn't it pushed to me yet? If documentation lags, willingness to pay drops immediately. I've said before that delayed and messy documentation damages customer confidence; this is even more obvious in car scenarios because users touch the steering wheel every day and won't wait for you to explain the architecture.

These past few days, I've been testing some things related to embodied robots, and the feeling is more intuitive. Car systems at least have mature whole-vehicle after-sales service, clear model boundaries, and road scenarios; many robots are still in high-end pilot phases, with toolchains and after-sales systems not yet running smoothly. Li Auto making the Sensory Cockpit a versioned delivery is paving the way for subsequent software services.

I don't focus much on feature lists; I mainly look at whether getting in the car is more worry-free, if parking has fewer mistakes, and if family trips have fewer complaints. If an OTA can turn navigation destination selection, route choice, and refueling decisions into a natural interaction, it has the chance to become part of optional packages or long-term services.

But there are boundaries here too. Cars are safety products; experience optimization cannot spill over into risk. The "Comfortable" and "Efficient" styles essentially hand decision-making power to the user, which also requires manufacturers to clearly define the boundaries. Which scenarios are usable, which require human takeover, and who takes responsibility if something goes wrong—these matter more than the buzzwords at launch events.