Low-Dimensional Motor Control Representations in Throwing Motions

Core muscle activity during suspension exercises

As Human-Robot Interaction (HRI) develops, robots are expected to learn more complex and demanding interaction skills. Complex HRI tasks are often embodied in robots that are jointly constrained by task space and joint space. In the field of Imitation Learning, scholars have explored the topic of joint constraints from task space and joint space, but there are few relevant studies in Human-Robot Interaction. In this paper, based on the Interaction Primitives framework (a HRI framework), we propose an interaction inference method that first generalizes the robot's movements in two spaces synchronously and then probabilistically fuses the two movements based on Bayesian estimation. This work was validated in the task that the robot follows a human handheld object, and the inference errors (RMSE and MAE) of the method are smaller in both task space and joint space than in IP using only singlespace inference.

It is well-known that a human musculo-skeletal body is redundant in terms of both kinematics and dynamics. The former means that the degree of freedom in joint space is larger than that in task space, and the latter means that a joint is driven by a number of muscles. All human skillful movements can be performed by using both redundancies. However, these redundancies induce the underlying ill-posedness problem that each joint angle and muscle’s output forces cannot be uniquely determined. These ill-posedness problems are known as “Bernstein’s problem” and are important to understand how human multi-joint movements are produced. In this study, we address the latter redundancy problem on how muscle’s output forces can be determined from the viewpoint of robotics. In this paper, we consider a reaching movement by means of a two-link planar arm with six muscles and show that both damping and elastic properties coming from nonlinear dynamics of the muscles play a crucial role. By using a simple task space feedback control input together with an additional term to control the internal force to regulate damping and elasticity in joint space, we show some simulation results which exhibit human-like quasi-straight line movement.

Control problems of motion of multi-body mechanical systems under constraints and/or with redundancy in system’s degrees-of-freedom (DOF) are treated from the standpoint of Riemannian geometry. A multi-joint reaching problem with excess DOF is tackled and it is shown that a task space PD feedback with damping shaping in joints maneuvers the endpoint of the robot arm to reach a given target in the sense of exponentially asymptotic convergence. An artificial potential inducing the position feedback in task space can be regarded as a Morse-Bott function introduced in Riemannian geometry, from which the Lagrange stability theorem can be directly extended to this redundant case. The speed of convergence of both the orbit of the endpoint in task space and the trajectory of joint vector in joint space can be adjusted by damping shaping and adequately choosing a single stiffness parameter. In the case that the endpoint is constrained on a hypersurface in E 3, the original Lagrange dynamics expressed in an implicit form by introducing a Lagrange multiplier is decomposed into two partial dynamics with the aid of decomposition of the tangent space into the image of the endpoint Jacobian matrix and the kernel orthogonally complemented to the image. The stability problem of point-to-point endpoint movement on the constraint surface is reduced to the former case without constraint.

Chapter 1 - Terminology and general definitions

This paper proposes a novel parallel robot - Virtual Ground Robot (VGR) - that is driven by three Series Elastic Actuators (SEAs) to interact with a human. The proposed Virtual Ground Robot provides a virtual ground on which a human can stand on and interact in three directions: the pitch, the roll and the height directions. The most significant features of the proposed VGR are that 1) it is driven by RFSEAs (Reaction Force-sensing Series Elastic Actuator), and thus it can provide precise forces and torques, 2) the size of the VGR is small enough for a human to stand on with ease, and 3) it can generate torque/force large to support a weight of a human. Taking advantage of RFSEAs utilized in the proposed VGR, Spatial Force control algorithm is proposed in this paper. In order to design this controller, the motions of VGR are defined in the task space, the joint space and the RFSEA level. Based on the Kinematics, force control of VGR in the task level, which is named Spatial Force Control is designed and verified using experiments.

TMS assessed cortical representation scales with stimulation intensity and muscle activation

The time-optimal path following (OPF) problem is to find a time evolution along a prescribed path in task space with shortest time duration. Numerical solution algorithms rely on an algorithm-specific (usually equidistant) sampling of the path parameter. This does not account for the dynamics in joint space, that is, the actual motion of the robot, however. Moreover, a well-known problem is that large joint velocities are obtained when approaching singularities, even for slow task space motions. This can be avoided by a sampling in joint space, where the path parameter is replaced by the arc length. Such discretization in task space leads to an adaptive refinement according to the nonlinear forward kinematics and guarantees bounded joint velocities. The adaptive refinement is also beneficial for the numerical solution of the problem. It is shown that this yields trajectories with improved continuity compared to an equidistant sampling. The OPF is reformulated as a second-order cone programming and solved numerically. The approach is demonstrated for a 6-DOF industrial robot following various paths in task space.

Previously, a stretching regimen was designed for manual material handling (MMH) of gas cylinders as a potential ergonomic solution for reducing occupational injury. No studies have made use of objective process measures, such as muscle activation levels, for evaluation of effects of stretching programs. Examine acute effects of stretching on muscle activation levels and driver perceived level of exertion in gas cylinder handling during simulated delivery operations. A within-subject experiment was conducted with eight male participants being subjected randomly to two conditions over a two-day period: stretching before delivery trials and no stretching. Surface electromyography and the Borg CR-10 scale for perceived exertion were used. Generally, results were variable among muscle responses. The extensor muscle bundle in the forearm was found to show a significant decrease (p = 0.0464) in activation level because of stretching. The anterior deltoid and trapezius significantly increased (p < .0001) the EMG activation level with stretching. Also counter to expectations, participants rated perceived exertion significantly higher (p = 0.0423) for trials preceded by stretching. This research indicates a muscle stretching regimen in advance of MMH activities has mixed effects on activation levels across muscles. It is possible that effects are attributable to body posture positions, or manner of muscle use, during actual work activities. Findings indicate that stretching prior to work activity does have an impact on specific muscle activation.

This paper reports an investigation of the magnitude and timing of the stretch reflex over the full range of activation of flexor carpi radialis. While it is well established that the magnitude of the reflex increases with the level of muscle activation, there have been few studies of reflex magnitude above 50% of maximum voluntary contraction (MVC) and virtually no study of the timing of the response in relation to activation level. Continuous small amplitude (approximately 2 degrees) perturbations were applied to the wrist of 12 normal subjects while they maintained contraction levels between 2.5-95% MVC, monitored via surface electromyography (EMG). Both narrow band (4-5 Hz) and broad band (0-10 Hz) stretch perturbations were employed. The gain (EMG output/stretch input) and phase advance of the reflex varied with the level of muscle activation in a similar manner for both types of stretch, but there were significant differences in the patterns of change due to stretch bandwidth. Consistent with previous studies, the group average reflex gain initially increased with muscle activation level and then saturated. Inspection of individual data, however, revealed that the gain reached a peak at about 60% MVC and then decreased at higher contraction levels, the pattern across the full range of activation being well described by quadratic functions (mean r2=0.82). This quadratic pattern has not been reported previously for the neural reflex response in any muscle but is consistent with the pattern that has been reliably observed in studies of the mechanical reflex response in lower limb muscles. In contrast to the pattern for reflex gain, the phase advance of the reflex (at a stretch frequency of 4.5 Hz) decreased linearly from approximately 130 degrees at the lowest contraction levels to approximately 50 degrees as maximum voluntary contraction was reached (mean r2=0.69). This decrease corresponds to a delay of 49 ms introduced centrally in reflex pathways. All subjects showed clearly defined quadratic functions relating reflex gain and linear functions relating reflex phase to activation level, but there were considerable individual differences in the slopes of these functions which point to systematic differences in synaptic behaviour of the motoneuron pool. Thus, there was wide inter-subject variation in both the contraction level at which the reflex gain reached a peak (31-69% MVC) and the highest target contraction level that could be sustained during reflex measurement (47-95% MVC). A high correlation between these variables (r2=0.78) suggests a linear relation between afferent support of contraction and muscle fatigability. The decline in reflex gain at high levels of muscle activation signals a failure of muscle afferent input and subjects in whom the gain reached a peak and declined early were unable to sustain higher target contraction levels. The results of the study show that both the timing and magnitude of the stretch reflex vary markedly over the full range of voluntary muscle activation. The pattern of variation may account for why the stretch reflex contributes most effectively to muscle mechanics over the lower half of the range of activation, while progressive reductions in both gain and phase advance at higher levels render the reflex mechanically less effective and make tremor more likely.

This work is concerned with the contouring control problem of robotic manipulators using the method of equivalent errors. The method was previously proposed for multi-axis motion systems with emphasis on the machine tools applications. It is shown in this paper that the method of equivalent errors is equally applicable to the robot manipulators. For robotic systems, a desired path is usually described in terms of the position of the end-effector (task space), but the system dynamics are described in generalized coordinates (joint space). The proposed method utilizes the forward kinematics to transform the desired path in the task space into that in the joint space. As such, the equivalent errors can be easily defined. A two-link robot manipulator contouring a circular path has been numerically studied. It is found that the contouring errors are virtually zero (order of fento meter), verifying the effectiveness of the method.

The design and control of a robot requires the computation of some mathematical models such as the transformation models between the joint space (in which the configuration of the robot is defined) and the task space (in which the location of the end-effector is specified). These transformation models are very important since robots are controlled in the joint space, whereas tasks are defined in the task space. The modeling of robots in a systematic and automatic way requires an adequate method for the description of their structure. Several methods and notations have been proposed. The most popular among these is the Denavit-Hartenberg method. This method is developed for serial structures and presents ambiguities when applied to robots with closed or tree chains. For this reason, we will use the notation of Khalil and Kleinfinger, which gives a unified description for all mechanical articulated systems with a minimum number of parameters.

We studied target reaching tasks involving not only the arms but also the trunk and legs, which necessitated some trunk flexion. Such tasks can be successfully completed using an infinite number of combinations of segment motions due to the inherent kinematic redundancy with the excessive degrees of freedom (DOFs). Sagittal plane motions of six segments (shank, thigh, pelvis, trunk, humerus, and forearm) and dynamic torques of six joints (ankle, knee, hip, lumbar, shoulder, and elbow) were analyzed separately by principal component (PC) analyses to determine if there was a commonality among the shapes of the respective waveforms. Additionally, PC analyses were used to probe for constraining relationships among the 1) relative magnitudes of segment excursions and 2) the peak-to-peak dynamic joint torques. In summary, at the kinematic level, the tasks are simplified by the use of a single common waveform for all segment excursions with 89.9% variance accounted for (VAF), but with less fixed relationships among the relative scaling of the magnitude of segment excursions (62.2% VAF). However, at the kinetic level, the time course of the dynamic joint torques are not well captured by a single waveform (72.7% VAF), but the tasks are simplified by relatively fixed relationships among the scaling of dynamic joint torque magnitudes across task conditions (94.7% VAF). Taken together, these results indicate that, while the effective DOFs in a multi-joint task are reduced differently at the kinematic and kinetic levels, they both contribute to simplifying the neural control of these tasks.

It is important to be caught a loading process during a care work with a unstable posture in worker's body quantitatively for its evaluation from dynamics viewpoint. Particularly, a waist part in trunk system often causes a disease because of the over load during a care work supporting a body of a patient. In order to clarify the mechanism of its over load, there are many evaluations by the joint moment method. However, the activation level of muscle against this over load is hardly estimated by joint moment method. A parameter such as strain of muscle to estimate its activation level is most effective. In this research, multi-body dynamics analysis of a waist part in consideration of muscle contraction during a care work is performed. Then, the activation level of muscle is estimated by the animation of motion of trunk system.

Chapter 3 - PID Control in Task Space