Archon Robotics Launches Anchor 1, a Humanoid Robot Whole-Body Control Model

Recently, Archon Robotics, an embodied intelligence enterprise located in Xuhui and incubated by Shanghai Chuangzhi Academy, officially launched the new generation humanoid Robot full-body control model Anchor 1, and simultaneously released relevant technical research and robot practical operation demonstration results to the public.

With the continuous integration of the artificial intelligence and robotics industries, embodied intelligence has become a hot track in current technological development. Humanoid robots are no longer satisfied with just simple walking movements; they also need to autonomously complete various operations in real and complex environments. But how to make robots coordinate full-body movements to stably and accurately complete various tasks has always been a major technical difficulty in the industry.

Anchor 1可以把运动跟踪误差降低30%～80% Anchor 1 can reduce motion tracking errors by 30%～80%

Anchor 1 was created specifically to solve this difficult problem. Its core idea is not to rigidly copy human limb movements, but to grasp the true action intentions behind the task.

Taking pushing a door as an example, a human's arm length, joint angles, and robot's are different, and joint movement trajectories cannot be directly copied. What is truly key is the route of the hand approaching the door handle, the change in the body's center of gravity, and the switching of the two-foot support status. Anchor 1 precisely grasps these core logics, combined with the motion control base model system S0, to convert task ideas into stable full-body actions of the robot, allowing the robot to adapt to its own mechanical structure and steadily complete various operations in the real world.

Actual test data shows that Anchor 1 can reduce motion tracking errors by 30%～80% and requires fewer front-end inputs. Under the same deployment process, teleoperation data collection efficiency is higher, and the success rate of the robot autonomously completing tasks is also significantly improved.

从动图可见Anchor 1在不同姿态和移动条件下的全身协调控制能力 From the GIF, the full-body coordinated control capability of Anchor 1 under different postures and movement conditions can be seen

To train this model, the team combined a large amount of real human motion data and simulation training, using over 400,000 segments of real human motion samples. This method can reduce the workload of extensive manual programming and point-by-point debugging under traditional modes, helping robots possess stronger general operation capabilities. Multiple practical tests show that, compared to existing similar models, Anchor 1's global task success rate is significantly improved, and it is capable of adapting to full-body collaborative work under various postures and moving scenes.

The implementation of Yuance Weilai's technical achievements this time is both an important breakthrough for the enterprise in the field of intelligent control of humanoid robots and a vivid practice of Shanghai Chuangzhi Academy in talent cultivation, achievement transformation, and enterprise incubation, adding another fresh case for Xuhui to cluster embodied intelligence innovation resources and strengthen new momentum for the artificial intelligence industry.