A team led by Jiaxin Huang has presented X-Nav, which it calls the first system to take a two-legged humanoid robot on a long mission from indoors to outdoors from a plain-language instruction. On a Lingxi X2 humanoid, X-Nav completed a mission of about 100 m from an office lobby to a shop in a neighbouring building, according to the team's technical report on arXiv.
Its navigation model, MELIOR, uses only images from the robot's built-in camera, with one version predicting actions and another a navigation goal, while LiDAR and odometry handle the final approach and low-level control. An agent workflow, X-Navarch, breaks a short human instruction into sub-tasks. The robot followed a change of destination, walked through a door from the lobby to the road and got past static and moving obstacles.
Training data comes from open-source scenes, digital twins of the deployment sites and real robots. During deployment, the model was adapted with data from a digital twin of the site, which raised its success rate in the twin. The report ties the work to Worken, AGIBOT's measure of productive robot work, which depends on robots reaching their workplaces on their own.
Background
X-Nav
Humanoid navigation system
System that takes a two-legged humanoid robot on long missions from indoors to outdoors from a plain-language instruction.
AGIBOT
Humanoid robot maker
Shanghai robotics company founded in 2023 by former Huawei engineers, building humanoid robots for industry and services.



