Future humanoid robots will be widely deployed in our daily lives. Motion planning and control in an unstructured, confined, and human-centered environment utilizing dexterity and a cooperative ability of dual-arm robots is still an open issue. We propose a globally guided dual-arm reactive motion controller (GGDRC) that combines the strengths of global planning and reactive methods. In this framework, a global planner module with a prospective task horizon provides feasible guidance in a Cartesian space, and a local reactive controller module addresses real-time collision avoidance and coordinated task constraints through the exploitation of dual-arm redundancy. GGDRC extends the start-of-the-art optimization-based reactive method for motion-restricted dynamic scenarios requiring dual-arm cooperation. We design a pick–handover–place task to compare the performances of these two methods. Results demonstrate that GGDRC exhibits accurate collision avoidance precision (5 mm) and a high success rate ($84.5\%$).

In this work, we tackle the challenge of generating human-like grasp motions on humanoid robots—a task that remains difficult for traditional methods due to the complexity of modeling human arm joint preferences and synthesizing dexterous grasps for novel objects. Our approach, termed anthropomorphic grasp motion planning (AGMP), integrates deep learning models by leveraging on human demonstrations and annotated grasps. Specifically, we learn a Riemannian metric in the joint space from human data to capture the directional preferences of upper limb movements, and formulate a quadratic Riemannian distance function as the optimization objective. To assess the grasp quality, we propose a dexterous grasp evaluator that accepts candidate grasps and object as inputs, returning scalar costs. By exploiting the differentiable nature of these networks, we iteratively solve the problem through approximations to both the learned Riemannian distance and the grasp constraint. Simulation results demonstrate that our approach reduces trajectory errors by up to 40% compared to the minimum-jerk model. Moreover, our approach improves the success rate by 11% and significantly reduces planning time relative to existing methods. We further validate our method on a humanoid robot prototype in realworld experiments, achieving a 73.3% success rate.

In humanoid robot manipulation imitation learning, arm and tool synergies are required to accomplish tasks. However, the existence of arm and tool shape variations within the demonstrators and between the demonstrator-robot impacts the generalization performance. This paper models the arm and tool as an extended kinematic chain, taking the elbow, wrist, tool end, and tool orientation as keypoints. We propose a humanoid arm manipulation generalization method based on keypoints synergetic guidance, including Postural and Movement Synergy. Using principal component dimensionality reduction and feature space optimization, Postural Synergy obtains the task's start and end positions corresponding to the new target under the new kinematic chain shape. Movement Synergy generates keypoints' trajectories adapted to the new kinematic chain through a dynamical system with coupling terms, and then obtains robot joint trajectories by optimization-based inverse kinematics. Experiments are conducted under shape variations of both the arms and tools. The results indicate that our method significantly reduces keypoints' errors in Postural Synergy compared to Tool-Centric, Motion Retargeting and Cartesian space optimization methods. In Movement Synergy, our method generates dynamically similar motions using coupling terms. Real robot generalization results show that our method is able to improve the task success rate remarkably while mimicking human motion at the same time.