whole-body control

This work addresses the challenge of achieving safe and compliant whole-body manipulation for legged robots in dynamic environments by proposing a hybrid architecture that integrates model-driven admittance control with a reinforcement learning–based gait policy. Safety during physical interaction is ensured through a reference governor, while a neural network–enhanced Kalman filter improves base velocity estimation accuracy. A unified whole-body force response is realized using six-degree-of-freedom force/torque sensing. Experimental validation on the Unitree Go2 platform demonstrates significant improvements in high-precision interaction tracking, compliant human–robot collaboration, and safety-critical reliability in dynamic scenarios.

This work addresses the lack of reusable, cross-platform compliant control infrastructure in existing robotic software, which hinders unified algorithm implementation and high-level interfacing. The authors propose a robot-agnostic, modular compliance control framework that decouples controller infrastructure from control laws via a plugin architecture. It supports variable impedance control in both joint and Cartesian spaces and, for the first time, enables online adaptation of the primary compliance direction according to task geometry—overcoming the limitations of fixed coordinate frames. Built upon the ROS ecosystem, the framework leverages Pinocchio to parse URDF models for kinematic and dynamic computations and employs runtime plugin loading with generic wrappers to interface heterogeneous hardware. Real-world and simulated experiments demonstrate significant performance improvements in contact-intensive tasks and seamless transferability across multiple robotic arms.

Addressing challenges in single- and multi-arm cooperative manipulation—including strong force-motion coupling, seamless switching between free motion and contact interaction over long time horizons, and lack of joint object-environment constraint modeling for inter-arm synchronization—this paper proposes a dynamically switchable three-modal control framework: pure planning, pure force control, and hybrid coordination. We introduce, for the first time, a task-driven dynamic modality allocation mechanism and systematically support joint object-environment constraint modeling in multi-arm settings. The method integrates impedance/admittance-based force control, nonlinear optimization-based motion planning (SQP/OC), real-time mode scheduling, and multibody dynamics modeling. Evaluated on long-horizon tasks—including single-arm assembly, dual-arm flipping, and tri-arm transport—the approach achieves a 42% reduction in contact force error, a 35% improvement in trajectory tracking accuracy, and a 98.7% task success rate.

针对人形机器人在物理交互中对全身控制的需求，提出LAC方法，通过合成数据集和强化学习训练策略实现线性和角顺应性。

To address the challenge of jointly ensuring motion stability and manipulation force control in legged mobile manipulation, this paper proposes a full-order inverse-dynamics-based whole-body model predictive control (MPC) framework. The method directly optimizes joint torques within a single prediction horizon, unifying whole-body motion planning and contact force generation to achieve dynamically consistent and constraint-complete natural coupling behavior. It integrates Pinocchio for rigid-body dynamics modeling, CasADi for automatic differentiation, and Fatrop for efficient interior-point optimization, enabling real-time control at 80 Hz on a Unitree B2 quadrupedal platform equipped with a Z1 manipulator. Experimental validation demonstrates robust performance across diverse dynamic manipulation tasks—including dragging heavy objects, pushing boxes, and wiping whiteboards—significantly enhancing both robustness and generalization capability of legged mobile manipulation.

研究提出了一种基于MPPI的任务空间控制框架，通过实时求解刚体动力学并利用扭矩采样控制架构实现高效并行计算，从而在非结构化环境中实现安全有效的机器人操作。

This work addresses the challenge of simultaneously ensuring balance and safety in floating-base robots during continuous physical human–robot interaction, where existing whole-body control methods often suffer from steady-state errors or inflexible joint-space allocation. The authors propose a three-layer control architecture: a centroidal model predictive controller plans contact forces; a priority-driven whole-body controller generates joint torques via contact-consistent nullspace projection; and a Kalman-enhanced moving horizon quadratic program suppresses interaction disturbances in the residual nullspace. A novel covariance inflation mechanism ensures continuity of disturbance estimation across contact switches, achieving zero steady-state error under bounded constant interaction forces. Furthermore, an impedance equivalence theorem is introduced, proving that adaptive task-space impedance behavior can be exactly recovered over an infinite horizon. The method is validated on a 17-DOF bipedal robot and the Unitree G1 humanoid, supporting real-time operation at ≥1 kHz with high precision and minimal steady-state error.

This study addresses the challenge of maintaining physically stable grasps with dexterous robotic hands under dynamic contacts, modeling errors, and external disturbances. To this end, it proposes a real-time force regulation framework that operates without tactile sensing. By fusing geometric estimation with proprioceptive feedback, the method dynamically computes whole-hand contact force distributions that satisfy friction constraints and actuator consistency. Integrated with reactive motion planning, this approach enables closed-loop control across grasp acquisition, maintenance, and post-failure re-grasping for a 27-degree-of-freedom arm-hand system. Simulation results demonstrate significantly enhanced disturbance rejection capabilities, while hardware experiments validate stable grasping under complex contact evolution and rapid recovery following human-induced perturbations.

研究通过结合自由形式和约束控制的混合控制框架，解决复杂全身机器人遥操作中的协调感知、双臂操作和导航问题，提高任务效率并减少关节限制风险。

Current robotic teleoperation systems struggle to provide force/torque feedback, limiting the deployment of active compliance strategies in complex domestic environments. To address this challenge, this work proposes UME—a low-cost, lightweight, and portable upper-limb exoskeleton that integrates real-time torque-based haptic feedback with full-body motion capture, enabling transparent teleoperation for blind manipulation of constrained objects for the first time. UME employs a universal retargeting algorithm to interface seamlessly with diverse heterogeneous robotic arms and leverages a hybrid learning framework combining imitation and reinforcement learning to train whole-body compliant control policies. Experimental results demonstrate high success rates across a range of challenging tasks, including long-duration mobile manipulation, force-controlled box flipping, visually occluded pushing, and dexterous operations in confined desktop spaces.