MEPM (Multi-modal Emotion Perception Model) builds on the JEPA direction and is purpose-built for emotion recognition, assessment, intervention and optimization.
Not a lightly fine-tuned general model, but a domain foundation rebuilt for emotion perception.
Optimized for emotion perception, not a quick fine-tune of a general model.
Perception and closed-loop strategies tuned for continuous companionship, rather than passive reaction.
Fusion of behavioral, physiological, semantic and environmental signals for holistic assessment.
Supports our own hardware as well as API / SDK licensing.
Not a replacement for LLMs or World Models, but the layer that tells them when to act and how to stay within bounds when facing a person
| Model Type | Primary Modeling Target | Typical Output | What's Still Missing When Used Alone |
|---|---|---|---|
| Conventional Affective Computing | Facial expression, voice and text cues | Emotion labels / scores | Whether appearance equals state, and whether to intervene |
| Large Language Model (LLM) | Language, knowledge, task reasoning | Dialogue, content, planning | Whether to speak right now, and which memories are trustworthy |
| World Model (WLM) | Space, objects, action transitions | Prediction, navigation, manipulation | Human internal state, interaction timing and boundaries |
| Emotion Perception Model (MEPM) | Multi-source human state, environmental causality and interaction history | State confidence, action strategy and memory governance | Not a substitute for expression or physical prediction, but the setter of human boundaries for both |
Solving "how not to bump into the table" — helping robots understand and act on the physical world.
Solving "how not to cross a person's boundaries" — MEPM is not a plug-in for world models, but the missing social-state layer in embodied devices.
A technology foundation linking lightweight consumer devices and professional senior-care robots, with scenarios rolled out in phases
Built as a "spherical sprite plush art toy" with lightweight emotion perception algorithms, XinXiaoMiao focuses on light everyday emotional companionship for all ages at home. Perceptive, interactive and companionable, it brings AI naturally into family life.
A one-piece round plush design with ears and a curled tail as its signature features; each of the four characters has its own personality, suiting different companionship preferences
Desktop Companionship and Family Health Management
Mobile Health Companion Robot
From underlying architecture and product vehicles to external delivery, building a proactive, precise and privacy-preserving emotional companionship system
Licensing emotion perception capabilities via API / SDK to empower smart hardware, senior-care platforms and smart homes.
Selling our own devices to B2C households and B2B institutions, spanning consumer-grade and professional-grade product lines.
Model capabilities are first validated on our own devices, then extended to a broader partner ecosystem, forming a data and service loop.
Model capabilities are continuously validated and iterated in real scenarios, forming a two-way data and service loop
View Larger
For care facilities and community service centers: group emotion monitoring, companionship services and health data archiving.
View Larger
In hospital wards, providing continuous companionship and emotional comfort, helping medical staff track changes in patients' emotional states.
View Larger
Working with OEMs and Tier 1 partners to embed emotion perception in the cockpit, adapting interaction and alerts to occupant state.
Platform Resources × Joint Research × Productization: Accelerating the Commercialization of Emotion Perception
Drawing on investment, scenarios, channels, policy and industrialization resources to build innovation ecosystem support.
Planned to be led by a CAS academician, jointly with the national technology innovation center and key university labs.
A desktop AI companion robot serving home and everyday health management scenarios.
A mobile health companion robot linking care for people living alone with professional senior-care services.
Model roadmap: HeartVigor 1.0 (multimodal emotion perception model) → HeartVigor 2.0, with software licensing opening in parallel
From fundamental algorithms to end applications, building a complete technology chain from foundation model R&D to consumer product delivery
Industry-specific AI model development, training and deployment, offering full-stack solutions from fundamental algorithms to industry applications, covering multimodal emotion perception models, domain-specific models and an AI Agent development platform.
Covering humanoid robots, embodied intelligence systems and core algorithm R&D; providing complete robots and intelligent upgrade solutions, spanning VLA manipulation algorithms, whole-machine design and AI upgrades for installed equipment.
Represented by AI art toys and companion robots, applying multimodal emotional interaction to consumer scenarios to create "soulful" smart desktop products and AI interactive terminals.
AI technology roadmap planning, architecture design and feasibility assessment
Integration and deployment of AI systems with existing enterprise IT / OT systems
Data processing, annotation and training data management
Building and operating a new-media content matrix across 8 major platforms
Whether it's API / SDK licensing, hardware ecosystem partnership, or deploying senior-care and home companionship scenarios, we welcome a conversation.
Contact Us