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

HUSKYLENS 2 is a next-generation AI vision sensor designed to make powerful machine vision more accessible for robotics, education, interactive systems, and intelligent device prototyping. Powered by a Kendryte K230 dual-core 1.6GHz processor with an integrated 6 TOPS AI accelerator, it runs vision algorithms directly on-device for fast, low-latency processing. It includes more than 20 ready-to-use AI models, supports the deployment of custom-trained YOLO models, and features a built-in MCP service that provides structured real-world perception data to compatible large language models and AI agents. Standard UART and I2C interfaces allow connection to Arduino, micro:bit, Raspberry Pi, ESP32, UNIHIKER, and other popular controllers.

This TF Card Version uses a pre-installed 8GB automotive-grade TF card as its system storage. The processor, 1GB LPDDR4 memory, 6 TOPS AI computing performance, built-in models, custom model support, recognition speed, and recognition accuracy remain unchanged from the eMMC version. HUSKYLENS 2 also features a replaceable 2MP camera module, a 2.4-inch IPS touchscreen, a microphone, speaker, fill lights, and an RGB indicator.

HUSKYLENS 2 TF Card Version

To maintain stable product availability amid limited eMMC supply, this version uses a pre-installed 8GB automotive-grade TF card as its system storage.

The TF card has been selected for high endurance, stable operation, and resistance to vibration, and has been validated for use with HUSKYLENS 2. The processor, DDR memory, 6 TOPS AI computing power, built-in models, custom model deployment, recognition speed, and recognition accuracy are identical to those of the standard eMMC version.

Blazing-Fast On-Device AI with 6 TOPS of Power

At the core of HUSKYLENS 2 lies a dual-core 1.6GHz processor with an integrated 6 TOPS AI acceleration unit. This substantial computing power, combined with 1GB of RAM, enables HUSKYLENS 2 to run complex algorithms and neural networks directly on-device. This facilitates real-time processing for demanding tasks like multi-object detection and intricate pose estimation, ensuring high-speed, low-latency performance without reliance on cloud computation.

Start Creating Instantly with Over 20 Built-in AI Models

HUSKYLENS 2 is designed for immediate productivity. It comes pre-loaded with a diverse library of more than 20 AI models, covering a wide array of common computer vision tasks from face recognition to object tracking. This extensive collection allows users to start building projects right out of the box with minimal setup, significantly lowering the barrier to entry for AI application development.

Empowering LLMs with Contextual Awareness via MCP Service

HUSKYLENS 2 is the first AI vision sensor to feature a built-in MCP service, fundamentally changing how LLM perceives the physical world. While other camera provide raw image "snapshots," HUSKYLENS 2 delivers a structured, intelligent brief. It can inform a Large Language Model (LLM) that it sees a specific, pre-identified person performing a particular action. This endows LLMs with accurate, context-aware perception, enabling them to make smarter, more informed decisions.

Deployment of Custom Models Trained with YOLO

Beyond the built-in models, users can train their own vision models using YOLO and deploy them on HUSKYLENS 2. HUSKYLENS 2 comes with a complete set of annotation, training, and conversion tools, making it easy to create custom-trained models.

Flexible Model Combination for Diverse Applications

HUSKYLENS 2 architecture supports the serial or parallel execution of multiple models. This capability enables the creation of more complex and nuanced applications. For instance, combining hand keypoint detection with object tracking can create sophisticated contactless control systems, where a gesture can initiate the tracking of a specific object. This multi-model approach allows for layered logic and more intelligent interactive behaviors.

Real-time Video Streaming for Monitoring and Debugging

HUSKYLENS 2 facilitates seamless data collection and image display through its real-time video streaming function. It supports both wired (USB-C) and wireless (via an optional Wi-Fi module) image transmission. This is critical for applications requiring remote monitoring, for collecting datasets to train new models, and for debugging visual applications by providing a live view of what the sensor sees.

Replaceable Camera Module for Any Application

HUSKYLENS 2 is equipped with a modular camera module, allowing it to easily adapt to a wide range of scenarios—from microscopic inspection to long-range monitoring. Its unmatched flexibility supports interchangeable lenses, including manual focus, microscope, and night vision modules.

Broad Compatibility with Mainstream Controllers

With standard UART and I2C interfaces, HUSKYLENS 2 ensures seamless connectivity with a wide range of microcontrollers and single-board computers, including Arduino, micro:bit, Raspberry Pi, ESP32, UNIHIKER K10, and UNIHIKER M10.

Master Every Feature with the Massive Tutorial Wiki

HUSKYLENS 2 is supported by an incredibly detailed tutorial wiki that covers every single feature. This resource provides step-by-step walkthroughs for all 20+ models, advanced parameter tuning, and specific code and graphical programming examples for Arduino, Raspberry Pi, micro:bit, and UNIHIKER.

Model Hub for Ready-to-Use Vertical Models (Coming Soon)

To further accelerate development, a Model Hub will be launched. This online platform will host a curated collection of official and user-generated models tailored for specific vertical markets, such as agriculture, retail, and safety monitoring. Users will be able to browse, download, and deploy these specialized models directly to their HUSKYLENS 2, providing instant access to powerful, niche AI capabilities.

Video: HUSKYLENS 2 AI Vision Sensor Introduction & Demos