Ching-An Cheng

Self-learning assistive exoskeleton with sliding mode admittance control

Human intention estimation is important for assistive lower limb exoskeleton, and the task is realized mostly by the dynamics model or the EMG model. Although the dynamics model offers better estimation, it fails when unmodeled disturbances come into the system, such as the ground reaction force. In contrast, the EMG model is non-stationary, and therefore the offline calibrated EMG model is not satisfactory for long-time operation. In this paper, we propose the self-learning scheme with the sliding mode admittance control to overcome the deficiency. In the swing phase, the dynamics model is used to estimate the intention while teaching the EMG model; in the consecutive swing phase, the taught EMG model is used alternatively. In consequence, the self-learning control scheme provides better estimations during the whole operation. In addition, the admittance interface and the sliding mode controller ensure robust performance. The control scheme is justified by the knee orthosis with the backdrivable spring torsion actuator, and the experimental results are prominent.

Bayesian human intention estimator for exoskeleton system

The estimation of the human applying torque is critical in many applications, especially in the design of assistive exoskeleton. The most common approaches are the estimation by the inverse dynamics or by the EMG signal. However, the EMG-based torque estimation is not always stable owing to the sweats of skin, the noise from posture change, and the nonlinear mapping between the EMG signal and the human torque. In addition, the estimation based on the dynamic model is unstable in the multi-DOFs system and especially in the existence of exogenous disturbance, such as ground reaction force. Therefore, we propose the Bayesian human intention estimator and the graphical model of human-exoskeleton system to solve these issues. Through the experiments, the proposed method can merge the information from both the EMG signal and dynamic model, and can make the estimated torque more stable.

Design of a new hybrid control and knee orthosis for human walking and rehabilitation

Simultaneously considering the physical interaction between the user and the robot within safety and performance constraints in rehabilitation and human walking situations, this paper proposes a new backdrivable torsion spring actuator (BTSA) with hybrid control that switches between direct electromyography (EMG) biofeedback control and zero impedance control, to provide a novel rehabilitation training and walking assistance mechanism for humans. The proposed backdrivable 1-DOF serial elastic actuator is designed to achieve intrinsic safety, compliance properties, and control performance. The proposed mechanical system can provide desirable backdrivable property and softer stiffness than that of traditional robots. In additional, the proposed hybrid control not only considers the assistive function, when human assistance is required, but also the compliance property, when assistance is not needed. Compared to state-of-the-art assistive methods, the BTSA with the proposed hybrid control system is unique in that it can simultaneously achieve assistance control through EMG biofeedback and compliance control through zero impedance control. A simple human-robot interaction model is built to investigate performance and explain the whole control concept. Further, a knee exoskeleton is built and three kinds of controls are used on a human subject to demonstrate the difference between them. Both simulation and experimental results show that the proposed BTSA mechanism with hybrid control offers the desired properties.

Virtual Impedance Control for Safe Human-Robot Interaction

Collision avoidance is essential for safe robot manipulation. Especially with humans around, robots should work only when safety can be robustly guaranteed. In this paper, we propose using virtual impedance control for reactive, smooth, and consistent collision avoidance that interferes minimally with the original task. The virtual impedance control operates in the risk space, a vector space describing the possibilities of all forthcoming collisions, and is designed to elude all risks in a consistent response in order to create assuring human-robot interaction experiences. The proposed scheme intrinsically handles kinematic singularity and the activation of avoidance using a boundary layer defined on the spectrum of Jacobian. In cooperation with the original controller, the proposed avoidance scheme provides a proof of convergence if the original controller is stable with and without projection. In simulations and experiments, we verified the characteristics of the proposed control scheme and integrated the system with Microsoft Kinect to monitor the workspace for real-time collision detection and avoidance. The results show that the proposed approach is suitable for robot operation with humans nearby.

Humanoid robot push-recovery strategy based on CMP criterion and angular momentum regulation

We propose a push-recovery strategy to stabilize the robot under unmodelled, large external forces. The strategy integrates Center-Of-Gravity (COG) angular momentum regulator, COG state estimator, and stepping control, which online modifies the trajectories of the COG and the swing leg. Using the centroidal-moment-pivot criterion, the COG angular momentum regulator controls the dynamics of the COG as an impedance system through the feedback of COG state estimator based on Kalman filter. The stepping control, on the other hand, selects the appropriate balancing reaction in anticipation of the potential consequences of the external disturbances on the robot. In simulations and experiments, we show the proposed push-recovery strategy can effectively save the robot from falling down and walk more smoothly.

Learning the Inverse Dynamics of Robotic Manipulators in Structured Reproducing Kernel Hilbert Space

We investigate the modeling of inverse dynamics without prior kinematic information for holonomic rigid-body robots. Despite success in compensating robot dynamics and friction, general inverse dynamics models are nontrivial. Rigid-body models are restrictive or inefficient; learning-based models are generalizable yet require large training data. The structured kernels address the dilemma by embedding the robot dynamics in reproducing kernel Hilbert space. The proposed kernels autonomously converge to rigid-body models but require fewer samples; with a semi-parametric framework that incorporates additional parametric basis for friction, the structured kernels can efficiently model general rigid-body robots. We tested the proposed scheme in simulations and experiments; the models that consider the structure of function space are more accurate.

Novel feature of the EEG based motor imagery BCI system: Degree of imagery

Motor imagery recognition has been considered an important topic in the brain-computer interface (BCI) community. Due to noises and artifacts in signals, how to gain satisfactory classification accuracy is still a critical issue. We propose in this paper a novel feature to address this issue. The method consists of three steps. Firstly, EEG signals from different electrodes are transformed by Time-Frequency Analysis method, in this paper Hilbert-Huang Transform. A set of features, Degree of Imagery (DOI) are then extracted from the spectrums by the proposed feature extraction method. The features can effectively represent the event-related-desynchronization (ERD) during motor imagery. Experimental results on the BCI 2003 competition dataset III indicate that our method achieves better classification accuracy and higher mutual information (MI) than other researches using the same dataset and with low computational time, which is capable of real-time usage.

Efficient grasp synthesis and control strategy for robot hand-arm system

This research is aimed to improve the efficiency of grasping with hand-arm system. It is divided into two parts: The first part proposes a simplified grasping condition for grasp planning and uses the tactile sensors at fingertip to control the interaction impedance for a robot hand to handle uncertainties in real-world applications. The second part extends the previous techniques from the robot hand to a full robot hand-arm system, and a four-stage planning is designed to generate an ideal grasp trajectory. We demonstrate the whole system by illustrating how our humanoid robot - NINO - with NTU-Hand V would open a door. By adopting this research in hand-arm coordination, mobile robots or humanoid robots can perform grasping and manipulating tasks to improve our daily lives.

Identification of the inverse dynamics of robot manipulators with the structured kernel

The inverse dynamics model of robots is often the key for accurate control. Especially in the computed torque control, the nonlinearity and the friction can be compensated, leading to better performance. The inverse models, however, is not trivial. The traditional Euler-Lagrange model based on the rigid body assumption often underfits in the presence of frictions and requires tedious derivations; the learning-based model needs larger training data set, since the structure of the dynamics is not considered. To overcome the aforementioned issues, we propose a structured kernel to replace the rigid body model and combine it with the universal radial basis kernel by direct sum. The proposed structured kernel asymptotically has the same convergence rate as the traditional model, and is general regardless of the configuration of the robot. Therefore, no analytic derivation is needed. Together with the universal radial basis kernel, the proposed approach enjoys the advantages of both the conventional and the learning-based models. To verify the proposed method, the simulations are used to investigate the performance in terms of the prediction errors.

Learn the Lagrangian: A Vector-Valued RKHS Approach to Identifying Lagrangian Systems

We study the modeling of Lagrangian systems with multiple degrees of freedom. Based on system dynamics, canonical parametric models require ad hoc derivations and sometimes simplification for a computable solution; on the other hand, due to the lack of prior knowledge in the systems structure, modern nonparametric models in machine learning face the curse of dimensionality, especially in learning large systems. In this paper, we bridge this gap by unifying the theories of Lagrangian systems and vector-valued reproducing kernel Hilbert space. We reformulate Lagrangian systems with kernels that embed the governing Euler-Lagrange equation-the Lagrangian kernels-and show that these kernels span a subspace capturing the Lagrangians projection as inverse dynamics. By such property, our model uses only inputs and outputs as in machine learning and inherits the structured form as in system dynamics, thereby removing the need for the mundane derivations for new systems as well as the generalization problem in learning from scratches. In effect, it learns the systems Lagrangian, a simpler task than directly learning the dynamics. To demonstrate, we applied the proposed kernel to identify the robot inverse dynamics in simulations and experiments. Our results present a competitive novel approach to identifying Lagrangian systems, despite using only inputs and outputs.