Monumental Labs has built the first end-to-end autonomous CNC stone carving system on KUKA KR240 robots, turning a 3D model and a marble block into a finished sculpture with no human in the loop. The team presented the work at Embodied AI NYC, combining machine learning, computational geometry and robot control for a concrete embodied-AI deployment in traditional craft.
Sim-to-Real
仿真到现实
Technique of transferring policies trained in simulated environments to physical reality
Curated Intelligence
Hongqi Chain has terminated its 16-month robot retail project after initial stock market hype, citing critical failures in handling dense shelves and irregular items. The collapse highlights the significant gap between lab-controlled environments and real-world open retail scenarios for embodied AI. Industry experts estimate a 2-5 year timeline before robots can reliably replace human staff, pending a 'GPT moment' for general manipulation capabilities.
Unitree founder Wang Xingxing revealed that the G1 humanoid robot has produced 10,000-20,000 units, making it the world's best-selling model. Tests show frontier LLMs like GPT-6 can directly control robots for tasks like painting and Go without training. The company is developing a self-evolving system called 'Unitree Preview', aiming for voice-controlled task completion by next year.
NVIDIA expanded its Isaac platform at GTC Berlin and partnered with Agile Robots to launch the Agile ONE SN, a humanoid reference platform pre-configured with Isaac GR00T and Jetson Thor. Scheduled for early 2027 release, it integrates simulation, training, and deployment tools to reduce setup time from weeks to days, targeting research and enterprise development.
Shanghai AI Lab and partners introduced Real2Gym, a framework that converts human or robot videos into interactive simulation environments using geometric constraints and physics engine validation. This approach addresses the gap between visual realism and physical feasibility in video reconstruction, enabling agents to learn skills in simulation and transfer them to real robots, resulting in a 74.9% reduction in token usage.
Hugging Face and NVIDIA released a guide for building healthcare robots using the Isaac platform, covering data collection in simulation, policy training, and deployment on real hardware. The tutorial aims to lower barriers for embodied AI in vertical domains by providing a reproducible end-to-end workflow.
OpenAI demonstrated robot controllers trained entirely in simulation that successfully deploy on physical robots to solve simple tasks. The key breakthrough is the shift from open-loop to closed-loop control, enabling robots to react to unplanned environmental changes in real time.
Researchers developed the world's first spam-detection AI trained entirely in simulation and successfully deployed on a physical robot. This achievement validates the feasibility of Sim-to-Real transfer learning for specific perception tasks, though the source text lacks detailed technical specifications or performance metrics.