Two provides an AI-driven social intelligence layer that can be licensed and integrated into existing humanoid robot platforms. The system stores encrypted user profiles and uses edge‑optimized multimodal deep‑learning to detect facial, vocal, and posture cues, adjusting speech, humor, and proximity in real time via a configurable policy engine. It is accessed through an SDK and REST API and supports GDPR‑compliant data handling with optional federated learning for model improvement.
Funding
Funding not disclosed
Founders
Product
Problem
Humanoid robots often lack the ability to remember individual user preferences, interpret subtle emotional cues, and adjust their behavior in real time, resulting in interactions that feel mechanical and can reduce user acceptance in home, care, and hospitality settings.
Solution
Two delivers an AI-powered social intelligence layer that can be integrated into existing humanoid robot platforms via a licensing model. The layer maintains persistent, encrypted user profiles that store preferences and interaction histories. It processes facial micro‑expressions, vocal prosody, and body posture using edge‑optimized deep‑learning models to infer the user's emotional state. Based on the inferred state, a policy engine dynamically adjusts speech cadence, humor level, and physical proximity to align with the user's needs. All inference runs locally on the robot’s compute hardware, minimizing latency and preserving privacy. Developers access the functionality through a documented SDK and RESTful API, enabling rapid incorporation into robot control stacks. The system also supports continuous improvement through optional, anonymized data aggregation for model retraining.
Target Audience
Primary customers are manufacturers of service‑oriented humanoid robots and research laboratories developing socially interactive robotic systems for residential care, assisted living, and hospitality environments.
Features
- Persistent user profile storage with preference recall and secure, encrypted data handling
- Multimodal emotion detection combining facial micro‑expression analysis, vocal tone parsing, and posture recognition
- Adaptive behavior engine that modulates speech rate, humor, and proximity based on real‑time emotional inference
- Edge‑optimized neural networks designed for low‑latency execution on typical robot onboard processors
- GDPR‑compliant data lifecycle management with role‑based access controls
- Comprehensive SDK and REST API for seamless integration into robot operating systems and middleware
- Configurable policy framework allowing developers to define custom response rules per application
- Optional federated learning pipeline that aggregates anonymized interaction data to refine models without exposing raw user information