Millimeter wave radar can measure distances, directions, and Doppler velocity for objects in harsh conditions such as fog. The 4D imaging radar with both vertical and horizontal data resembling an image can also measure objects' height. Previous studies have used 3D radars for ego-motion estimation. But few methods leveraged the rich data of imaging radars, and they usually omitted the mapping aspect, thus leading to inferior odometry accuracy. This paper presents a real-time imaging radar inertial odometry and mapping method, iRIOM, based on the submap concept. To deal with moving objects and multipath reflections, we use the graduated non-convexity method to robustly and efficiently estimate ego-velocity from a single scan. To measure the agreement between sparse non-repetitive radar scan points and submap points, the distribution-to-multi-distribution distance for matches is adopted. The ego-velocity, scan-to-submap matches are fused with the 6D inertial data by an iterative extended Kalman filter to get the platform's 3D position and orientation. A loop closure module is also developed to curb the odometry module's drift. To our knowledge, iRIOM based on the two modules is the first 4D radar inertial SLAM system. On our and third-party data, we show iRIOM's favorable odometry accuracy and mapping consistency against the FastLIO-SLAM and the EKFRIO. Also, the ablation study reveal the benefit of inertial data versus the constant velocity model, and scan-to-submap matching versus scan-to-scan matching.

This letter describes an approach to achieve well-known Chinese cooking art stir-fry on a bimanual robot system. Stir-fry requires a sequence of highly dynamic coordinated movements, which is usually difficult to learn for a chef, let alone transfer to robots. In this letter, we define a canonical stir-fry movement, and then propose a decoupled framework for learning this deformable object manipulation from human demonstration. First, dualarms of the robot are decoupled into different roles (a leader and follower) and learned with classical and neural network based methods separately, then the bimanual task is transformed into a coordination problem. To obtain general bimanual coordination, we secondly propose a Graph and Transformer based model— Structured-Transformer , to capture the spatio-temporal relationship between dual-arm movements. Finally, by adding visual feedback of contents deformation, our framework can adjust the movements automatically to achieve the desired stir-fry effect. We verify the framework by a simulator and deploy it on a real bimanual Panda robot system. The experimental results validate our framework can realize the bimanual robot stir-fry motion and have the potential to extend to other deformable objects with bimanual coordination.

Coordinate calibration accuracy is a prerequisite for the achievement of collaborative motion of dual robots and visual positioning. In the case of AXB=YCZ, the associated coordinate relationships of hand-eye (X), robot-robot (Y), tool-flange (Z) need to be determined in order to provide accurate data for dual-robot manipulation. The classical independent estimations and separable solving of rotational and translational components inevitably cause certain inaccuracy and inconsistency. To this end, the work presented herein describes an efficient approach based on dual quaternion (DQ) to calibrate the hand-eye (X), robot-robot (Y), tool-flange (Z) transformations, corresponding to the dual-robot collaborative motion. After establishing the closed-form kinematic equation of dual robots in DQ form, the calibration process is divided into two steps: 1) the hand-eye transformation (X) calibration is performed by letting one robot with a visual sensor move, while keeping the other robot with end-tool in a fixed configuration, thereby allowing the calibration result to be obtained by solving the DQ-based linear equations with the SVD algorithm; 2) the simultaneous calibration of robotrobot (Y) and tool-flange (Z) is transformed into a DQ-based equation, and the results are obtained in a similar way. A simulation analysis is carried out, accounting for the presence of noise levels and different data pairs, while calibration experiments are also conducted to verify the accuracy and efficacy of the proposed method.

Upper limb exoskeletons show promise for improving functionalities required for stroke patients. Despite recent progress, most of current upper limb rehabilitation devices are still bulky, heavy, and less compliant to be applied. This article presents a cable-driven compliant upper limb rehabilitation exoskeleton robot (CURER) with a lightweight frame and comfortable human–robot interaction. A modular series elastic actuator (SEA) was designed to provide controlled torque for each active robotic joint, and Bowden cables were applied to transfer controlled torque to distal joints. A six-bar double parallelogram mechanism was designed to implement 3 active degrees of freedom (DOFs) of a shoulder. An actuated elbow with 1 DOF and a wrist with a passive DOF were also developed for CURER. The anthropomorphic shoulder, elbow, and wrist joints can minimize misalignment between human upper limbs and the robot. The length of anthropomorphic arm was adjustable for a wide range of users. It can apply up to a 33 N·m torque in shoulder flexion/extension, abduction/adduction, intra/extra rotation, and elbow flexion/extension, with a range of 7.6–8.0 Hz position bandwidth in each actuation. CURER has a large range of motion and can provide accurate torque control for stroke patients’ requirements. Besides, a comprehensive rehabilitation strategy including robot-in-charge mode and human-in-charge mode was developed for different recovery stages. Experiments carried out on CURER actuation units demonstrated good position and impedance control performance. Finally, a virtual reality training system was developed to assist the subjects to accomplish upper limb rehabilitation efficiently.

Many human manipulation skills are force relevant, such as opening a bottle cap and assembling furniture. However, it is still a difficult task to endow a robot with these skills, which largely is due to the complexity of the representation and planning of these skills. This paper presents a learning‐based approach of transferring force‐relevant skills from human demonstration to a robot. First, the force‐relevant skill is encapsulated as a statistical model where the key parameters are learned from the demonstrated data (motion, force). Second, based on the learned skill model, a task planner is devised which specifies the motion and/or the force profile for a given manipulation task. Finally, the learned skill model is further integrated with an adaptive controller that offers task‐consistent force adaptation during online executions. The effectiveness of the proposed approach is validated with two experiments, i.e., an object polishing task and a peg‐in‐hole assembly.

The development of soft hands is an important progress to empower robotic grasping with passive compliance while greatly decreasing the complexity of control. Despite the advances during the past decades, it is still not clear how to design optimal hands or fingers given the task requirements. In this paper, we propose a framework to learn the optimal design parameter for a fin-ray finger in order to achieve stable grasping. First, the pseudo-kinematics of the soft finger is learned in simulation. Second, the task constraints are encoded as a combination of desired grasping force and the empirical grasping quality function in terms of winding number. Finally, the effectiveness of the proposed approach is validated with experiments in simulation and using real-world examples as well.