SpaceMind - iFlytek's intelligent space agentic architecture

What is SpaceMind?

SpaceMind is a smart space agent architecture launched by iFlytek, upgrading spaces into L2.5 stage proactive intelligent agents with perception, understanding, memory, decision-making, and execution capabilities. The system employs millimeter-wave radar privacy awareness and a dual-routing architecture, achieving device response within 700ms and separating complex reasoning. Through collaboration between local semantic models and cloud-based large models, it enables spaces to proactively provide services based on people's location, state, and habits. SpaceMind has already been implemented in high-end residential and hotel scenarios in the Middle East, Latin America, and Southeast Asia.

SpaceMind's main functions

• Voice controlWith a single sentence, you can precisely adjust the color and brightness of the lights, the opening and closing of the curtains, and the playback of music.
• Visual object findingThe system uses a camera to locate the lost remote control and marks its exact location on the screen.
• Semantic scene linkageIt understands complex semantics such as "a cozy atmosphere is suitable for chatting" and synchronizes the lighting, curtains, screens, and speakers.
• Long-term preference memoryIt automatically captures users' preferences for rest, movie watching, music, and other scenarios, and proactively restores an immersive environment.
• Unobtrusive proactive protectionIt senses when someone gets up at night and automatically turns on soft path lights to avoid disturbing family members with strong light.
• Multi-protocol compatibilityNative support for Matter, Thread, and mainstream ecosystems such as Apple Home, Google Home, and Amazon Alexa.

SpaceMind's technical principles

• Millimeter-wave radar privacy awarenessSpaceMind uses millimeter-wave radar as its core sensing source. It does not collect images or record faces, but can achieve three-dimensional spatial positioning and breathing-level micro-motion detection with an accuracy of 5 centimeters. Even in complex environments such as darkness and smoke, it can continuously identify the number of people, distinguish between people and pets, and determine states such as stillness, movement, and falls, thus resolving the contradiction between privacy protection and reliable sensing.
• Dual-router decoupling architectureThe system adopts a separate design for the primary control channel and the secondary inference channel. The primary channel executes device control commands locally directly, with model inference taking approximately 200 milliseconds. The overall end-to-end response time is controlled within 700 milliseconds, improving speed by 100% and stability by 30%. The secondary channel handles complex semantic understanding, task planning, and agent tasks through a large cloud model, balancing real-time performance and deep inference.
• Dual-model collaborative systemThe local family semantic model is trained based on real family data, adopts native multilingual design and is fine-tuned by SFT supervision, and is responsible for understanding high-frequency simple commands and calling device functions; the cloud-based inference model undertakes complex tasks that require context and planning judgment, such as long-term memory, schedule management, conflict detection and proactive reminders.
• Multi-Agent Collaboration and Persistent MemoryMultiple agents understand the room status, manage lighting, control air conditioning, or handle schedules, and work together to achieve user goals. The system continuously accumulates information about family members, spatial structure, scene preferences, and long-term interaction history, gradually forming a long-term understanding of the family space and realizing a shift from passive response to proactive prediction.

How to use SpaceMind

• Hardware accessConnect your existing KNX, Zigbee, Matter, and other devices to the SpaceMind network using Gateway Lite or Gateway Pro without replacing your existing hardware.
• Perception DeploymentInstall millimeter-wave radar sensors and NOVA series native hardware to build a spatial perception layer covering the entire house.
• Voice wake-upAfter waking up the system, it issues natural language commands to quickly control devices such as adjusting lights, opening and closing curtains, or playing music.
• Scene ExperienceWhen a user requests a complex semantic request such as "make the space more cozy," the system automatically coordinates and adjusts the lighting, curtains, screens, and audio systems accordingly.
• Seamless serviceWithout requiring manual commands, the system proactively executes services based on perception, memory, and multi-agent collaboration, such as automatically turning on soft path lights when getting up at night.

SpaceMind's core advantages

• Privacy and securityMillimeter-wave radar solutions can achieve accurate perception without the need for cameras, completely avoiding visual privacy leaks.
• Ultra-fast responseThe dual-routing architecture keeps the end-to-end latency of device control within 700ms, improving stability by 30%.
• Active intelligenceBased on persistent memory and state awareness, the space can proactively provide predictive services before the user speaks.
• Multi-Agent CollaborationBreak down device silos and allow all devices in the house to be uniformly scheduled according to human intentions, rather than operating independently.
• Global Eco-CompatibilitySupports multi-protocol and multi-platform access, lowering the barrier to global deployment.

SpaceMind's Competitive Comparison

Application scenarios of SpaceMind

• High-end residencesThe whole house actively senses the environment and automatically adjusts the lighting, curtains, and sound to provide personalized scene services.
• Smart HotelCross-regional automated room management creates an immersive check-in and seamless interactive experience for guests.
• Commercial spaceBy linking the KNX art panel with intelligent lighting, intelligent aesthetic control of the exhibition hall and office environment can be achieved.
• Elderly careUtilizing millimeter-wave radar for non-visual, privacy-protecting detection of falls and abnormal conditions, providing timely warnings and safeguarding safety.
• Child companionshipIt senses when a child gets up at night and automatically turns on soft path lighting to avoid disturbing family members with strong light and noise.