AI Learns to Act
AI startup DeepMirror says it has integrated the OpenClaw framework into its Physical AI stack. The company claims the move will help bridge the critical "reasoning-to-action" gap in robotics, allowing AI models to more effectively translate digital instructions into physical manipulation.
The open-source framework at the center of the integration, OpenClaw, was created by Austrian developer Peter Steinberger. It moves beyond chatbot-like responses by giving large language models (LLMs) the ability to run on a local machine, execute shell commands, modify files, and control hardware directly. In robotics, this architecture often uses ROS 2, the industry-standard framework. OpenClaw can act as a bridge, translating high-level goals expressed in natural language into concrete commands within the ROS 2 stack, effectively serving as an autonomous developer and system maintainer. This approach targets a core problem where AI's digital plans fail when they meet physical reality's perception errors and safety constraints. The integration aims to create verifiable robotic "skills" that allow a machine to autonomously choose its next action, confirm success, and recover from plans that don't match the real world. DeepMirror itself is a University of Cambridge spin-out that initially built its AI cloud platform for drug discovery and analyzing biomedical data. The company's expansion into robotics is managed through its dedicated Physical AI brand, Looper Robotics. The system's efficiency has been demonstrated on resource-constrained hardware; in one test, OpenClaw autonomously built and deployed hardware-connected ROS applications on a Raspberry Pi 4. This highlights its potential for on-device maintenance and monitoring in real-world, not just datacenter, environments. Recent breakthroughs using the framework show its potential for creating "world memory" in robots. By integrating data from LiDAR and cameras on a Unitree humanoid robot, OpenClaw has enabled the machine to recall people, objects, and past events, giving it an understanding of space and time.
