Motion intention detection is fundamental in the implementation of human-machine interfaces applied to assistive robots. In this paper, multiple machine learning techniques have been explored for creating upper limb motion prediction models, which generally depend on three factors: the signals collected from the user (such as kinematic or physiological), the extracted features and the selected algorithm. We explore the use of different features extracted from various signals when used to train multiple algorithms for the prediction of elbow flexion angle trajectories. The accuracy of the prediction was evaluated based on the mean velocity and peak amplitude of the trajectory, which are sufficient to fully define it. Results show that prediction accuracy when using solely physiological signals is low, however, when kinematic signals are included, it is largely improved. This suggests kinematic signals provide a reliable source of information for predicting elbow trajectories. Different models were trained using 10 algorithms. Regularization algorithms performed well in all conditions, whereas neural networks performed better when the most important features are selected. The extensive analysis provided in this study can be consulted to aid in the development of accurate upper limb motion intention detection models.

The lack of intuitive and active human–robot interaction makes it difficult to use upper-limb-assistive devices. In this paper, we propose a novel learning-based controller that intuitively uses onset motion to predict the desired end-point position for an assistive robot. A multi-modal sensing system comprising inertial measurement units (IMUs), electromyographic (EMG) sensors, and mechanomyography (MMG) sensors was implemented. This system was used to acquire kinematic and physiological signals during reaching and placing tasks performed by five healthy subjects. The onset motion data of each motion trial were extracted to input into traditional regression models and deep learning models for training and testing. The models can predict the position of the hand in planar space, which is the reference position for low-level position controllers. The results show that using IMU sensor with the proposed prediction model is sufficient for motion intention detection, which can provide almost the same prediction performance compared with adding EMG or MMG. Additionally, recurrent neural network (RNN)-based models can predict target positions over a short onset time window for reaching motions and are suitable for predicting targets over a longer horizon for placing tasks. This study’s detailed analysis can improve the usability of the assistive/rehabilitation robots.

Conventional grippers fall behind their human counterparts as they do not have integrated sensing capabilities. Piezoresistive and capacitive sensors are popular choices because of their design and sensitivity, but they cannot measure pressure and slip simultaneously. It is imperative to measure slip and pressure concurrently. We demonstrate a dual slip-pressure sensor based on a thermal approach. The sensor comprises two concentric microfabricated heaters maintained at constant temperature. An elastic dome, with embedded liquid metal droplets, is placed on top of concentric heaters. Heat transfer between sensor and the object in contact occurs through the elastic dome. This heat transfer causes changes in the power absorbed by the sensor to maintain its temperature and allows for measurement of pressure while identifying slip events. Liquid metal droplets contribute to enhanced thermal conductivity (0.37 W/m-K) and reduced specific heat (0.86 kJ/kg-K) of the polymer without compromising on mechanical properties (Young's modulus—0.5 MPa). For pressure monitoring, sensor measures change in power ratio against increase in applied force, demonstrating a highly linear performance, with a high sensitivity of 0.0356 N ⁻¹ (pressure only) and 0.0189 N ⁻¹ (slip with simultaneous pressure applied). The sensor discriminates between different contact types with a 96% accuracy. Response time of the sensor (60–75 ms) matches the measured response time in human skin. The sensor does not get affected by mechanical vibrations paving way for easy integration with robotic manipulators and prosthetics.

We developed a compact 3-D end-effector type of upper limb assistive robot, named CARE Robot, that provides transparency movement and adaptive arm support control to achieve home-based therapy and training in the real environment. The proposed system composes five degrees of freedom, including three active motors and two passive joints at the end-effector module. The core structure of the system is based on a parallel mechanism. The key innovation of this work lies in developing a novel adaptive gravity compensation framework, uniquely designed to provide 3-D arm support without reliance on external sensors. Unlike conventional methods that require wearable magnetic or IMU sensors, our approach estimates joint angles through a fixed torso model and sagittal plane model, enabling personalized and efficient support force generation. To validate the feasibility of the proposed method, preliminary experiments with four healthy subjects were conducted, demonstrating significant reductions in muscle activity, highlighting the potential of our system to assist in Activities of Daily Living and home-based rehabilitation. The combination of the compact design and sensor-free joint estimation approach sets this work apart, offering a more accessible and efficient robotic solution for upper limb assistive therapy.

We developed a 3D end-effector type of upper limb assistive robot, named as Assistive Robotic Arm Extender (ARAE), that provides transparency movement and adaptive arm support control to achieve home-based therapy and training in the real environment. The proposed system composes five degrees of freedom, including three active motors and two passive joints at the end-effector module. The core structure of the system is based on a parallel mechanism. The kinematic and dynamic modeling are illustrated in detail. The proposed adaptive arm support control framework calculates the compensated force based on the estimated human arm posture in 3D space. It firstly estimates human arm joint angles using two proposed methods: fixed torso and sagittal plane models without using external sensors such as IMUs, magnetic sensors, or depth cameras. The experiments were carried out to evaluate the performance of the two proposed angle estimation methods. Then, the estimated human joint angles were input into the human upper limb dynamics model to derive the required support force generated by the robot. The muscular activities were measured to evaluate the effects of the proposed framework. The obvious reduction of muscular activities was exhibited when participants were tested with the ARAE under an adaptive arm gravity compensation control framework. The overall results suggest that the ARAE system, when combined with the proposed control framework, has the potential to offer adaptive arm support. This integration could enable effective training with Activities of Daily Living (ADLs) and interaction with real environments.