News
A Chinese research team has proposed the world’s first “force–position hybrid control algorithm.”
Release time: 2025-10-20
Recently, a Chinese research team has achieved a major breakthrough in the field of robotic algorithms, proposing the world’s first unified theoretical framework for a “force–position hybrid control algorithm.” This algorithm enables robots to simultaneously learn both position and force control without relying on force sensors, boosting the success rate of relevant tasks by approximately 39.5% compared with strategies that use only position control. The corresponding paper was awarded the Outstanding Paper Prize at the Conference on Robot Learning (CoRL), marking the first time since the prize’s inception that it has been bestowed upon an all‑Chinese research team.
Researchers note that the vision–language–motion models widely used today often fall short when tackling real-world tasks, primarily because these tasks typically involve complex contact scenarios. For example, when wiping a blackboard, a robotic arm must both conform to the surface and maintain appropriate pressure; when opening or closing a cabinet door, it must accurately sense the internal push‑pull spring mechanism. Robots need not only to know “where to go” and “where to reach,” but also to understand “how much force to apply.” Before the advent of hybrid force–position control algorithms, such requirements could only be addressed through force sensors.
The algorithm introduces UniFP, a unified theoretical framework that employs reinforcement learning to train a policy for estimating forces from the robot’s historical states and leverages position‑and‑velocity adjustments for compensation, thereby simulating a variety of positional commands, force inputs, and external disturbances. This policy enables diverse operational behaviors, including position tracking, force application, force tracking, and compliant interaction. Furthermore, the contact information incorporated into the force estimation module significantly enhances the performance of trajectory‑based imitation learning; in four contact‑rich manipulation tasks, the success rate is approximately 39.5% higher than that of policies relying solely on position control.
The core inspiration for this algorithm stems from impedance control: it treats the interaction between the robot’s end-effector and its environment as a spring–damper–mass system, regulating both position and force by controlling the resulting deviation. In practical operation, the robot no longer simply “mechanically follows a prescribed trajectory”; instead, it can sense the environment and actively apply forces to respond to real-time changes. Experimental results demonstrate that, in a blackboard‑cleaning task, conventional position‑control strategies either fail to achieve thorough cleaning or exert excessive force, whereas the new algorithm maintains a stable contact pressure and ensures the board is completely cleaned. In a complex scenario where a drawer is obstructed, the force‑sensing‑based approach boosts the success rate from 0.3 to 0.76.
The groundbreaking significance of this research lies in the fact that it represents the first control algorithm in the field of legged robotics capable of seamlessly integrating force and position control within a single framework. Not only does it reduce the hardware reliance on force sensors during robot deployment, but more importantly, it offers a new technological pathway for compliant manipulation in contact-rich environments. Industry experts believe that this algorithmic achievement is of great importance for enhancing the fundamental control capabilities of force‑controlled robots, providing more efficient technical support for the deeper integration of force‑control technologies into industrial applications such as precision assembly and flexible polishing.
Previous
Tel.：0512-68127800
E-mail：youngme@youngmeii.com
Address: F1-2, Building 12, No.38 Beiguandu Road, Yuexi Street, Wuzhong District, Suzhou City
©️ 2022 Jiangsu Younme IOT Intelligent System Co., Ltd. All rights reserved Powered by www.300.cn SEO
