# Real-Time Autonomous Systems with Edge AI on Raspberry Pi Run high-performance AI models on limited resources using LiteRT and Gemma. - **Source:** [Google Developers Blog](https://brifing.si/source/1e914d4a-9164-4a1a-82cb-3d169fbad877) - **Topic:** [Data & AI](https://brifing.si/c/data-ai) - **Original article:** - **On Brifing:** https://brifing.si/s/f92a0eef-aadd-4c2a-8be5-c756cd13f02f ## The Power of Edge AI Edge AI enables real-time artificial intelligence on local devices without cloud dependency. Compact devices like Raspberry Pi can run autonomous systems with low latency while maintaining data privacy. LiteRT provides an optimized runtime for such systems, supporting a wide range of models from classical ML to the latest LLMs right out of the box. ## Advantages of LiteRT Developed by Google AI Edge, LiteRT simplifies high-performance inference on Raspberry Pi. It offers optimized memory usage and execution for both CPU and GPU, utilizing the device's full computing potential. With cross-platform compatibility, developers can easily deploy their models. The LiteRT-LM layer ensures efficient operation of Gemma models on Raspberry Pi. ## Gemma Models Google's family of lightweight open models, Gemma, is ideal for resource-constrained devices like Raspberry Pi. Designed for autonomous agents, smart cameras, and social robotics applications, these models can execute complex multi-step workflows locally. The Gemma 4 E2B model achieves a generation speed of ~27.3 characters per second thanks to its efficient tokenizer. ## Hardware Performance The Raspberry Pi 5's quad-core ARM Cortex-A76 CPU delivers ~153.6 GFLOPS (FP32) processing power. The integrated Broadcom VideoCore VII GPU operates at 800 MHz, enabling parallel processing. With LiteRT's WebGPU (Vulkan) support, computer vision and audio models can run on the GPU. This leaves the CPU free for intensive tasks like LLM inference. ## Real-World Application Example The Reachy Mini robot uses LiteRT and Gemma on Raspberry Pi 5 to perceive its environment and respond in real time. The system distributes vision and language models between CPU and GPU for low-latency performance. The Gemma 4 E2B model achieves 99 tokens/sec for prefill and 9 tokens/sec for decode while using only 1432 MB of memory. ## Developer Tools LiteRT provides a comprehensive toolset for conversion, quantization, benchmarking, and inference. The LiteRT CLI tool allows developers to manage complex edge AI workflows with a single command set. These tools, integrable with AI coding agents like Google Antigravity, enable autonomous execution of multi-stage machine learning workflows. ## Resource Optimization LiteRT offers a distribution specifically optimized for IoT devices with low storage and memory footprint. Unlike traditional AI runtimes, it contains no desktop or server dependencies. This modular design guarantees efficient operation on resource-constrained devices like Raspberry Pi. The LiteRT community provides many ready-to-use models. ## The Future of Edge AI Technologies like LiteRT and Gemma are making edge AI more accessible. Powerful models running on affordable devices like Raspberry Pi accelerate the adoption of autonomous systems. Advantages like data privacy, low latency, and cloud independence make edge AI indispensable for industrial applications. Developers can now deploy complex AI solutions with ease. ## Sources covering this story - https://developers.googleblog.com/mastering-edge-ai-on-raspberry-pi-with-litert-and-gemma/ — --- _Summary written by Brifing from the original reporting. Copyright in the original article belongs to its publisher; read the full piece at the link above._