PelicanVLM
Pelican-VL (PelicanVLM) is an open-source embodied multimodal AI brain model developed by X-Humanoid (Beijing Innovation Center of Humanoid Robotics, China). Currently the largest-scale open-source embodied brain model with parameter scales from 7B to 72B, it supports three robotic embodiments: wheeled humanoid, bipedal humanoid, and industrial manipulator. The model handles VLA (Vision-Language-Action), whole-body control, and motion planning, with DPPO cyclic training for self-correction. Trained on a 4+ million sample dataset.
Primary use case: Embodied AI brain model for controlling humanoid and manipulator robots across grasping, navigation, and manipulation tasks
Height
170 cm
Weight
55 kg
DOF
20
Specifications
Physical
Height
170 cm
Weight
55 kg
Degrees of Freedom
20
Locomotion
unknown
Balance
N/A
Manipulation
Hand Type
unknown
Power & Autonomy
Autonomy Level
unknown
AI & Compute
AI Capabilities
vision, learning
Commercial
Availability
open_source
Primary Use Case
Embodied AI brain model for controlling humanoid and manipulator robots across grasping, navigation, and manipulation tasks
Related news
From robot olympics to factory floor
X-Humanoid Sets Out Its Roadmap for a Unified Embodied Model and Self-Evolving Robot Policies
X-Humanoid presented a roadmap at ROSCon China 2026 for a unified embodied model and self-evolving robot policies, with the 'Pelican' project positioned as a step toward unifying a world model and simulator. The announcement frames the company's vision for a common foundation that supports continuously improving robot policies.
“We want to turn high performance into high reliability”: The engineer behind China’s fastest humanoid robot
Jack Guo, a 34-year-old engineer at X-Humanoid (Beijing Humanoid Robot Innovation Centre), led the TianGong Ultra humanoid robot to a 100-meter sprint time of 8.64 seconds at the World Humanoid Robot Games, beating Usain Bolt's world record by nearly a second. Now Guo is focused on the harder challenge of turning high performance into reliability and usability for real-world applications, likening today's humanoids to the earliest cars that first need to be controllable before pairing reliable bodies with smarter AI brains.
TianGong Ultra: What’s Next for the Chinese Humanoid Robot That Outran Usain Bolt?
TianGong Ultra, a humanoid robot developed by Beijing's X-Humanoid, stunned the world by sprinting 100 metres in 8.64 seconds at the World Humanoid Robot Games, beating Usain Bolt's 9.58-second human record under different conditions. The company's engineers, led by Jack Guo, are now focused on making the robot reliable and useful for real-world applications like factory work and emergency rescue, rather than just achieving speed records.