Robotec.ai, a leading deep-tech company specializing in advanced simulation solutions for the testing and deployment of robotics solutions across key global industries, has collaborated with AMD and Liquid AI to demonstrate the first fully autonomous warehouse robot powered exclusively by AMD Ryzen™ AI processors.

The robot leverages Agentic AI capabilities to dynamically plan and execute tasks in real time without reliance on hard-coded scripts. Powered by Liquid AI's next-generation LFM2 Vision Language Models, it seamlessly combines perception, reasoning, and natural language understanding to interpret commands, detect safety hazards such as spills or blocked exits, and autonomously execute corrective actions. The collaboration unlocks the full potential of the platform, thanks to extensive testing in simulated environments. Simulation, created with Open 3D Engine, supports validation of embedded AI on real hardware, avoiding the costs and risks of physical testing. It is a step towards the future of reasoning robots that will intelligently respond to the changing environment around them.

Physical intelligence as a glimpse into the future of industrial robotics

Physical intelligence will create immense value through efficient platforms like mobile manipulators, combined with agentic AI that integrates foundation models and state-of-the-art, reliable robotics.

With this new demonstration, a flexible mobile robot, powered by agentic AI that is running on AMD silicon, operates within a warehouse with mixed traffic. It completes tasks specified by humans using natural language, and adapts to changing conditions through replanning. The robot also serves as an inspection agent, alerting operators whenever unexpected occurrences or safety issues are detected in the warehouse area.

Liquid VLM (LFM2-VL): Multimodal Intelligence for Embodied Autonomy

At the core of this system is Liquid AI’s LFM2-VL, a next-generation Vision Language Model designed for embedded, real-time intelligence. Compact yet powerful, it integrates perception, reasoning, and language understanding into a single multimodal foundation model optimized for AMD hardware. To tailor the model for agentic robotics downstream tasks, Liquid leveraged simulation-derived synthetic data provided by Robotec, enabling domain-specific fine-tuning and robustness in complex industrial environments.
LFM2-VL interprets visual scenes, performs context-aware reasoning, and plans goal-driven actions entirely on-device — eliminating latency and cloud dependency. Its efficiency and responsiveness enable the robot to operate safely and autonomously in dynamic industrial environments.

ROS 2-powered hardware-in-the-loop simulation for rapid OEM prototyping

The Robotec.ai team has created a hardware-in-the-loop (HiL) simulation running a ROS 2 stack and AI inference on a single Ryzen device. In this configuration, the simulation runs on a separate AMD-powered computer and delivers a virtual environment that is indistinguishable from the physical world to the robot. The HiL interface then connects the robot’s sensor and actuator signals directly to the simulator’s inputs and outputs, allowing the same control logic to be applied under realistic, reproducible conditions. The robot used in the simulation, RB-KAIROS+, is designed and produced by Robotnik.

It supports the transition of the HiL setup to the real-world setup, enabling rapid OEM prototyping and demonstrating the solution's robustness. In the long run, HiL accelerates the innovation cycles, significantly speeding up the R&D phase. Robotec.ai has collaborated with AMD on hardware-in-the-loop before, boosting performance with simulation-driven testing.

Human-in-the-loop: orchestrating robots with a UI on an industrial tablet

The platform comes with an easy-to-use tablet UI that displays the robot’s plan, missions, and reasoning steps, alongside live maps, camera feeds, and anomaly reports, which lets human operators monitor robotics operations in real-time and address potential issues as they arise. Seamless human-robot interface keeps humans in the loop to supervise workflows.

AMD platform: an excellent match for Physical AI

The AMD processor proves to be an excellent “brain” for agentic AI while also capable of running the robotics software stack in parallel. It is fast, compact, and efficient, with incredible performance in both speed and power efficiency, as measured by transparent metrics such as tokens, latency, and throughput. Initially demonstrated on an AMD Ryzen processor, Robotec.ai plans to transition to an AMD embedded x86 solution in the near future.

Robotec.ai, AMD and Liquid AI collaborate on Embedded Agentic AI at ROSCon 2025

AMD, Robotec.ai and Liquid AI will showcase Agentic AI in robotics at ROSCon 2025 in Singapore, where participants will have a unique chance to see the demo live at the AMD stand (Booth 17/18). Robotec.ai’s team will be present onsite to introduce the solution to the broader audience.

Authors:

• Adam Dąbrowski, CTO at Robotec.ai
• Maciej Majek, ML Engineer – Project Lead at Robotec.ai
• Tomas Thoresen, Industrial System Architect at AMD

Interested in our latest projects? Follow us on LinkedIn to always stay updated!
You can also visit our GitHub page to learn more about our open-source projects.