Summary
Robots that move around and use their arms to interact with objects often need to feel how much force they apply. But some robots don’t have special sensors to measure these forces directly, making control difficult. The authors developed a new learning method that estimates these forces without needing sensors, combining knowledge of physics with a small neural network to improve accuracy. This estimate helps the robot better control its arm and wheels together, allowing it to push, pull, and manipulate objects smoothly. They tested their method both in computer simulations and on a real robot doing tasks like opening doors and rotating valves.
What this means in practice
- •For robotics system integrators: Implement force-aware control in mobile robots without installing end-effector force sensors to improve grasping and manipulation tasks.
- •For industrial automation engineers: Equip wheeled-legged robots with sensorless force control to perform precise valve turning and surface wiping, reducing hardware complexity and maintenance.
Authors
Xuanqi Zeng, Jiaming Wang, Tianlin Zhang, Lingwei Zhang, Botian Xu, Weipeng Xia, Zhongyu Li, Yun-Hui Liu
Abstract
Force-controlled loco-manipulation requires a whole-body policy to coordinate locomotion and arm motion while regulating end-effector interaction forces. This is challenging under floating-base dynamics and changing support contacts, particularly when end-effector force/torque sensing is unavailable for control. This paper presents a force-aware reinforcement learning approach with hybrid sensorless force estimation for wheeled-legged loco-manipulation. The proposed method provides a structured estimate of the end-effector force as an explicit policy observation, enabling force-guided contact behavior without using an end-effector force/torque sensor for control. The force estimate is obtained by combining generalized momentum observation, contact-constrained wrench projection, and temporal residual learning: the model-based components extract the physically structured part of the whole-body disturbance, while the residual network compensates the remaining motion-dependent bias. The estimated force is integrated into a mode-conditioned whole-body policy with an axis-wise force/position selector, allowing free-space motion, pure force regulation, and hybrid force/position control within one controller. Simulation results demonstrate improved sensorless force estimation and force-control performance. Hardware experiments further validate the proposed controller through quantitative valve-rotation and hybrid wiping evaluations, together with force-guided door opening and zero-force human-guided motion on a real wheeled-legged platform.