Gravity: AI Vision Gesture & Posture Sensor

Gravity: AI Vision Gesture & Posture Sensor is an offline AI visual interaction module designed for gesture recognition, posture recognition, and custom static action learning. Integrating hand keypoint detection, human skeleton detection, and fixed gesture classification, this compact Gravity sensor enables natural motion-based control without cloud processing. With I2C/UART communication, host computer configuration, Arduino, ESP32, and micro:bit compatibility, and a 37×37mm footprint, this edge AI sensing board supports privacy-conscious embedded interaction projects, smart devices, robots, and classroom demonstrations. Fast deployment.

Three Recognition Modes for Flexible Interaction

The AI vision gesture sensor provides three operating modes that can be switched through host computer software or communication commands. Hand keypoint detection tracks 21 hand keypoints at approximately 5 FPS for close-range gesture interaction. Human skeleton detection tracks 17 body keypoints and supports multi-person simultaneous recognition at approximately 10 FPS. Fixed gesture classification includes 13 built-in common gestures, making quick validation possible immediately after power-up.

Figure: Gravity AI vision sensor

Custom Static Action Learning with Local Storage

Using the host computer software, the module can learn, name, manage, and delete custom static gestures and postures. Up to 8 gestures and 8 postures can be stored independently in local flash memory and retained after power loss. Once learning is complete, the visual interaction sensor can operate independently from the PC and output recognition results to a main controller for personalized triggers.

Figure: ustom Static Action Learning with Local Storage

Offline AI Inference for Privacy-Sensitive Projects

All AI inference runs locally on the module, requiring no network connection and no cloud upload of recognition data. This makes the offline posture recognition board suitable for privacy-sensitive smart home control, human-robot interaction, interactive installations, and educational prototypes. Local processing also simplifies deployment in environments where network access is unavailable or unreliable. This keeps the module practical for fast prototyping, classroom demonstrations, and privacy-conscious interactive products and demos.

Low Power Gravity Interface Integration

The compact 37×37mm PCB supports 3.3V~5V supply, I2C/UART communication, a Gravity interface, and a Type-C debug/configuration port. Typical operating current is approximately [email protected], while sleep current can be as low as 37μ[email protected] in low-power mode. A WAKEUP pin enables intermittent battery-powered designs and small embedded installations. This keeps the module practical for fast prototyping, classroom demonstrations, and privacy-conscious interactive products.

Figure: ustom Static Action Learning with Local Storage

Ideal for touchless smart home control, human-robot interaction, interactive art installations, gesture-controlled music boxes, interactive lighting, projection triggers, virtual character action triggering, STEAM education, and maker projects, this edge AI vision module gives developers a compact path to motion-based control while keeping recognition data local and private. This keeps the module practical for fast prototyping, classroom demonstrations, and privacy-conscious interactive products.