When large models enter the physical world: Xiaomi plans to add "memory" to smart homes

```

Smart home technology is moving from "understanding a sentence" to "remembering a home."

On June 18, Xiaomi officially released the "Whole House Intelligent AI Open Source Solution" Xiaomi Miloco 2.0.

This solution uses Xiaomi's self-developed MiMo large model as the intelligent core, and upgrades the interaction methods, product features, and memory system based on last year's Miloco 1.0.

Miloco 2.0 mainly integrates with OpenClaw via Agent plugins, supporting macOS, Linux, and Windows systems.

The most notable change is Xiaomi's introduction of a household memory AI system for the first time.

In the past, the core capability of smart home systems was mainly device linking. Users pre-set rules, and the system executes accordingly, such as turning on the living room lights at 7 PM, turning on the air conditioner when the temperature exceeds 28℃, or the user saying "turn off the lights" and the devices respond.

But this kind of smart home is essentially still "rule-based control." It can carry out clear instructions, but has difficulty understanding long-term living habits formed in a home, and cannot continuously adjust services according to the preferences of different family members.

Miloco 2.0 is attempting to address exactly this gap.

According to Xiaomi, Miloco 2.0 can remember the identities, preferences, routines, and habits of different family members and store long-term memories at the household level. It organizes observed patterns at dawn every day, turning repeated behaviors into long-term archives. For example, if it detects that a family member comes home late, the system proactively offers overtime care.

To support this household memory, Miloco 2.0 uses Mi Home cameras as the entry point for comprehensive modality sensing, combined with microphones, Mi Home devices, and the large model's capabilities, to continuously comprehend household scenarios. The system can use facial and body information to identify family members; if it can't recognize them immediately, they enter a "stranger pool" and await user confirmation before registration.

This actually points to a core issue for large AI models entering household scenarios: Without long-term memory, the models are still limited to one-time Q&A and single command executions, which is obviously no different from previous smart home's "rule-based control."

From a broader technical trend, the significance of Miloco 2.0 lies in providing a sample for observing large models moving into the physical world.

Previously, large models mainly operated in chatboxes, search bars, and office software, with interactions mostly confined to the screen.

But household scenarios are real physical environments. AI must continuously sense the environment, identify family members, understand behavioral changes, and mobilize real devices when necessary in order to be effective here.

In this sense, household memory is precisely the capability that large models need to fill when entering the physical world.

Risk disclosure and disclaimerThe market has risks, investment needs caution. This article does not constitute personal investment advice, nor does it take into account individual users’ specific investment goals, financial situation, or needs. Users should consider whether any opinions, views, or conclusions in this article are suitable for their particular situation. Invest accordingly, at your own risk. ```