AI is rapidly moving into the physical world, giving rise to physical AI. This shift enables systems to perceive, decide, and act in real time, interacting directly with dynamic, unpredictable environments. Robotics and commercial drones are at the forefront of this evolution, driving new levels of automation and efficiency across healthcare, retail, transportation, and manufacturing.

In high-risk environments such as agriculture, warehouses, and industrial plants, robots and drones reduce human exposure to hazardous conditions. Ground-based robots handle repetitive or physically demanding tasks, while drones provide visibility across large or hard-to-reach areas, enabling real-time monitoring, inspection, and data collection.

In commercial settings, robots are reshaping the customer experience. For example, in retail stores, robots can answer questions, fetch products, and provide information. Drones extend these capabilities beyond the store, supporting inventory tracking, last-mile delivery, and site-to-site logistics with speed and efficiency. Together, these systems create a more responsive, intelligent operating environment, where physical tasks and real-time insights work in sync.

Bringing these intelligent, autonomous systems to life requires advanced hardware, a rich software ecosystem, and state-of-the-art AI models.

Hardware: Enabling real-time intelligence and durability

At the hardware level, robotics and drones demand powerful, efficient compute capable of handling real-time perception, decision-making, and multimodal processing. While many deployments will leverage a hybrid approach that combines edge and cloud, processing at the edge remains critical for delivering the speed, responsiveness, and low latency these systems require.

MediaTek’s Genio Pro 5100 is designed to meet these demands for physical AI applications. Based on TSMC’s 3nm process and featuring our 8th generation NPU, it delivers over 50 TOPS of system-level generative AI acceleration. This enables unrivaled acceleration for traditional AI tasks such as machine vision and object classification.

For generative AI applications, Genio Pro 5100 supports token generation rates of up to 23 tokens per second for large language models (LLMs) with up to 7B parameters. Genio Pro 5100 enables advanced camera sensor fusion for vision-based applications, supporting up to 16 cameras. It also delivers rich user interfaces, supporting up to three 4K displays.

Deploying physical AI in robotics and drones requires systems that can operate reliably under a variety of real-world conditions, from extreme temperatures to demanding industrial settings. Genio Pro 5100 is engineered to meet these requirements, with an industrial operating temperature range of -40°C to +105°C.

For improved manufacturability and system reliability, Genio Pro 5100 supports side-by-side LPDDR5x. Additionally, the platform offers flexible system expansion through PCIe Gen 4, USB, dual 2.5GbE, and a wide range of standard interfaces. Backed by a guaranteed 10-year supply commitment, the platform safeguards long-term product roadmaps and reduces supply chain risk for industrial and robotics deployments.

Software: Unlocking developer flexibility

Software is what enables developers to translate AI models into real-world, deployable physical AI applications. Genio Pro 5100 supports multiple Linux operating systems, including Yocto, Debian, and Ubuntu distributions. It also supports industry-standard AI frameworks, including PyTorch, ONNX Runtime, and TensorFlow Lite, enabling seamless model development and deployment across a wide range of use cases.

For robotics applications, support for the open-source ROS 2 (Robot Operating System) framework helps streamline development, integration, and system orchestration. MediaTek’s robust developer resources further accelerate time-to-market, enabling developers to build, customize, and scale physical AI applications more efficiently.

AI models: From perception to action

Advances in AI models are expanding what robotics and drones can do in different environments. While large language models (LLMs) enable natural interaction, newer approaches such as video-language models (VLMs) and vision-language-action models (VLAs) make it possible for robotics and drone applications to go beyond observing and interpreting their environments to acting on those observations in real time.

These capabilities are enabling more autonomous systems, where robotics and drones can interpret their environments and respond intelligently and safely. Genio Pro 5100 is designed to support these emerging model architectures, providing the performance needed to run them efficiently.

Advancing physical AI

As physical AI moves from concept to deployment, Genio Pro 5100 provides the high-performance silicon, software flexibility, and AI model support needed for robotics and drones to operate intelligently. From edge inference and multimodal perception to autonomous decision-making, Genio Pro 5100 is powering the next wave of physical AI innovation.