Chinese hotel robot handles laundry from collection to delivery in just 18.6 seconds

China-based ACE Robotics has demonstrated a hotel laundry workflow using its Mundo world model, combining robotic skills with cloud-based planning to automate the process from collection to delivery. The system enables a robot to collect laundry baskets from guest rooms, transport them to a laundry room, operate washing machines and dryers, and return clean clothes. Mundo creates interactive digital twins of hotel environments from photographs, allowing robots to simulate tasks and learn how objects behave before operating in the real world. According to the company, a self-improving skills system uses feedback from real-world failures to refine performance, bringing the complete laundry workflow down to 18.6 seconds. Robots master laundry The Mundo system allows a robot to collect laundry baskets from guest rooms, transport them to a laundry room, operate washing machines and dryers, and return the clean clothes. The demonstration marks the first long-chain commercial application of Mundo since ACE Robotics and Nanyang Technological University’s S-Lab introduced the world model last month. At the center of the system is Mundo, which converts photographs of hotel rooms, corridors, and laundry facilities into interactive digital twins. Unlike conventional visual simulations, these environments incorporate physical properties such as object mass and friction, allowing robots to learn how objects respond to contact and movement. Each simulation produces data on robot trajectories, contact states, and visual feedback. This information helps refine the robot’s action policies before it attempts the same tasks in a real hotel environment. ACE Robotics uses a dual-model architecture to divide complex planning from real-time execution. A general-purpose cloud model, identified by QbitAI as GPT-Astra, handles task decomposition, reasoning, and strategy exploration inside Mundo. Once the system identifies a workable approach, it stores the procedure as a reusable robotic skill. A separate model running locally on the robot then retrieves and executes these skills to perform individual actions. The division is intended to avoid the delays associated with relying on cloud computing for every movement. ACE Robotics said its cloud-based approach took more than 30 minutes to complete a full laundry run, while the local execution model reduced the time to 18.648 seconds. A scheduling system called BotAgent coordinates the workflow, dividing laundry operations into 53 atomic actions and 14 higher-level semantic steps. It can arrange more than 200 possible action sequences based on factors such as room locations, elevator availability, and queues at washing machines. Robots learn from failure The system also incorporates a feedback loop designed to improve robotic skills through repeated real-world attempts. When an operation fails, the cloud model examines multi-camera footage and sensor logs to identify what went wrong. It can then propose changes to the robot’s movements or select alternative camera views to overcome visibility problems. In one example, the system switched to an overhead camera after the robot’s wrist-mounted camera became obstructed. In another, it modified the gripper’s lifting trajectory and fingertip angles after jeans became caught inside a dryer drum. The robot can also reset the environment autonomously between attempts. A separate model evaluates each round, allowing the system to retain better-performing skill versions and roll back less effective changes. For dryer unloading alone, ACE Robotics reported conducting 165 real-robot trials over 57 hours. These produced 37 improvement proposals and approximately 80 skill versions. The company said unloading cycles eventually settled at 170 to 210 seconds. The average proportion of jeans left inside the dryer fell from 48 percent to 32 percent, with the best recorded run reaching 19 percent. ACE Robotics identified instant retail as another potential application for the technology. However, the reported performance figures come from the company and have not been independently verified. Get the latest in engineering, tech, space & science - delivered daily to your inbox.Jijo is an automotive and business journalist based in India. Armed with a BA in History (Honors) from St. Stephen's College, Delhi University, and a PG diploma in Journalism from the Indian Institute of Mass Communication, Delhi, he has worked for news agencies, national newspapers, and automotive magazines. In his spare time, he likes to go off-roading, engage in political discourse, travel, and teach languages.