Gravity AI Pose/Gesture Recognition Sensor

Introduction

The DFRobot SEN0670 Gravity AI Pose/Gesture Recognition Sensor is an offline AI vision module for recognising user-defined full-body poses and hand gestures. Rather than requiring collected data or AI model training, it uses Windows configuration software to learn and store actions locally, making it a focused option for gesture-driven human-machine interfaces.

What it does

The SEN0670 can be taught up to eight custom hand gestures and eight custom full-body poses. Once learned, these actions are stored locally and reported to a host system through I²C or UART. USB is also provided as an interface option.

For pose tracking, the module detects 17 body keypoints at approximately 10 FPS. Hand recognition detects 21 keypoints at approximately 5 FPS. This combination supports projects that need to distinguish between predefined body positions and hand-based commands without sending recognition data to an external service.

The module operates from a 3.3 V to 5 V supply and measures 37 mm × 37 mm.

Applications

• Touch-free control interfaces for embedded devices
• Gesture-activated interactive displays or installations
• Pose-triggered robotics and automation demonstrations
• Educational projects exploring local AI vision and keypoint recognition
• Custom command inputs for human-machine interaction prototypes

Design considerations

• Custom gestures and poses are configured using Windows software, so system setup should account for access to a compatible Windows environment.
• Recognition results can be integrated over I²C or UART; select the interface that matches the available host-controller peripherals and software architecture.
• The 3.3 V to 5 V operating-supply range can simplify integration with common microcontroller platforms, but supply and logic-level requirements should still be verified for the complete system.
• The stated detection rates differ for body and hand keypoints: approximately 10 FPS for 17 body keypoints and approximately 5 FPS for 21 hand keypoints. Designs should account for these update rates when defining interaction timing.
• Local storage is limited to up to eight learned gestures and eight learned full-body poses.

Availability and further information

Further product information is available from the Mouser Electronics source link and the DFRobot manufacturer link on this page.