Gravity: HUSKYLENS 2 - 6 TOPS LLM MCP AI Vision Sensor (20+ Models, Deploy Custom Models)

The Gravity: HUSKYLENS 2 is a next-generation edge AI vision sensor powered by a dual-core 1.6GHz Kendryte K230 processor delivering 6 TOPS of computing power. Designed for makers, students, and robotics developers, it features over 20 built-in vision models, custom YOLO model deployment, and integrated Model Context Protocol (MCP) support for LLMs. With a 2.4-inch IPS touchscreen, onboard audio, and broad microcontroller compatibility, it brings fast on-device vision to intelligent interactive projects.

Dual-core processor with 6 TOPS compute

Driven by a dual-core 1.6GHz K230 processor and 1GB of LPDDR4 RAM, the sensor delivers 6 TOPS of on-device AI acceleration. This enables high-speed, low-latency execution of complex neural networks, multi-object detection, and pose estimation entirely offline without cloud dependencies.

Over 20 pre-loaded AI models

The sensor includes an extensive library of more than 20 built-in vision algorithms, spanning face recognition, object tracking, line following, OCR, and hand keypoint detection. These pre-configured models allow users to start prototyping functional AI vision projects straight out of the box.

Contextual LLM awareness via MCP

HUSKYLENS 2 integrates a built-in MCP (Model Context Protocol) service that translates visual data into structured summaries for Large Language Models. Rather than sending raw images, it provides context-rich perception data about detected subjects and actions to enable informed autonomous decisions.

Custom YOLO model training and deployment

Users can train custom vision models using YOLO and deploy them directly onto the hardware using the provided annotation, training, and conversion toolset. This allows makers to create and run bespoke neural networks tailored to specific classification and detection tasks.

Serial and parallel multi-model execution

The processing architecture supports running multiple AI algorithms in parallel or serial pipelines for layered project logic. For example, pairing gesture detection with object tracking facilitates sophisticated contactless interfaces and reactive robotic behaviours.

Modular replaceable camera unit

The device features a 2MP 60FPS sensor on an interchangeable modular mount that easily adapts to specialised applications. The hardware supports swappable lenses and modules, including manual focus, night vision, and microscope units.

Broad microcontroller and SBC compatibility

Standard Gravity UART and I2C interfaces provide plug-and-play communication with Arduino, micro:bit, Raspberry Pi, ESP32, and UNIHIKER boards. Real-time video feeds can also be streamed via USB-C or an optional slot-in Wi-Fi module for live monitoring and dataset collection.

Video

Downloads & documentation

• Product wiki (DFRobot)
• KC