Stance Tasks Research Articles

Research on Cloud-Edge-End Collaborative Computing Offloading Strategy in the Internet of Vehicles Based on the M-TSA Algorithm

In the Internet of Vehicles scenario, the in-vehicle terminal cannot meet the requirements of computing tasks in terms of delay and energy consumption; the introduction of cloud computing and MEC is an effective way to solve the above problem. The in-vehicle terminal requires a high task processing delay, and due to the high delay of cloud computing to upload computing tasks to the cloud, the MEC server has limited computing resources, which will increase the task processing delay when there are more tasks. To solve the above problems, a vehicle computing network based on cloud-edge-end collaborative computing is proposed, in which cloud servers, edge servers, service vehicles, and task vehicles themselves can provide computing services. A model of the cloud-edge-end collaborative computing system for the Internet of Vehicles is constructed, and a computational offloading strategy problem is given. Then, a computational offloading strategy based on the M-TSA algorithm and combined with task prioritization and computational offloading node prediction is proposed. Finally, comparative experiments are conducted under task instances simulating real road vehicle conditions to demonstrate the superiority of our network, where our offloading strategy significantly improves the utility of task offloading and reduces offloading delay and energy consumption.

Principal component analysis of whole-body kinematics using markerless motion capture during static balance tasks

Balance tests have clinical utility in identifying balance deficits and supporting recommendations for appropriate treatments. Motion capture technology can be used to measure whole-body kinematics during balance tasks, but to date the high technical and financial costs have limited uptake of traditional marker-based motion capture systems for clinical applications. Markerless motion capture technology using standard video cameras has the potential to provide whole-body kinematic assessments with clinically accessible technology. Our aim was to quantify poses and movement strategies during static balance tasks (tandem stance, single limb stance, standing hip abduction, and quiet standing on foam with eyes closed) using video-based markerless motion capture software (Theia3D) and principal component analysis to examine the associations with age, body mass index (BMI) and sex. In 30 healthy adults, the mean poses for all balance tasks had at least one principal component (PC) that differed significantly by sex. Age was significantly associated with the PC describing leg height for the hip abduction task and erect posture for the quiet standing task. BMI was significantly associated with the PC capturing knee flexion in the single leg stance task. The movement strategies used to maintain balance showed significant differences by sex for the tandem stance pose. BMI was correlated with PCs for movement strategies for hip abduction and quiet standing tasks. Results from this study demonstrate how markerless motion capture technology could be used to augment analyses of balance both in the clinic and in the field.

Super Intendo: Semantic Robot Programming from Multiple Demonstrations for taskable robots

When an end-user instructs a taskable robot on a new task, it is important for the robot to learn the user’s intention for the task. Knowing the user’s intention, represented as desired goal conditions, allows the robot to generalize across variations of the learned task seen at execution time. However, it has proven challenging to learn goal conditions due to the large, noisy, and complex space of goal conditions expressed by human users. This paper introduces Semantic Robot Programming with Multiple Demonstrations (SRP-MD) to learn a generative model of latent end-user task goal conditions from multiple end-user demonstrations in a shared workspace. By learning a generative model of the goal conditions, SRP-MD generalizes to task instances even when the quantity of objects to be arranged is not in the training set or novel object instances are included. At test time, a new goal is pulled from the learned generative model given the objects present in the initial scene. The efficacy of SRP-MD as a step toward taskable robots is shown on a Fetch robot learning and executing bin packing tasks in a simulated environment with grocery items.

The Differences in Transient Characteristics of Postural Control between Young and Older Adults across Four Different Postural Tasks

Recently, the transient characteristics of postural sway have been suggested as an alternative or an improvement to traditional whole-trial analyses, possibly reflecting the sensory reweighing process. The purpose of this study was to assess if the transient characteristics of postural sway are sensitive to age and postural tasks. Twenty young adults (18–27 years old) and fifteen older adults (67–80 years old) performed parallel (eyes open and eyes closed), tandem and single-leg postural tasks for 30 s. Centre of pressure (CoP) velocity, amplitude and frequency were derived from force plate data. In addition to whole-trial estimates, we calculated the relative differences between the 1st and the 2nd (DIF_21) and the 1st and the 3rd (DIF_31) 10 s time intervals. Postural sway increased with the difficulty of the task, and the increase was more pronounced in the older adult group (p < 0.001; η2 = 0.31–0.38 for group × task interactions). Large and statistically significant age × task interactions were shown for both DIF_21 and DIF_31 for CoP anterior–posterior frequency (p < 0.001; η2 = 0.17–0.18). Possible age differences were also indicated for the transient response in CoP medial–lateral velocity in the parallel stance task. Specifically, we found an exaggerated transient response (i.e., relatively higher COP velocity in the first 10 s) in older adults compared to young adults when visual information was restored during the parallel quiet stance. This study shows the potential for an application of measures of the transient behavior of postural sway after the removal or restoration of visual information.

Mission specification and decomposition for multi-robot systems

Service robots are increasingly being used to perform missions comprising dangerous or tedious tasks previously executed by humans. However, their users—who know the environment and requirements for these missions—have limited or no robotics experience. As such, they often find the process of allocating concrete tasks to each robot within a multi-robot system (MRS) very challenging. Our paper introduces a framework for Multi-Robot mission Specification and decomposition (MutRoSe) that simplifies and automates key activities of this process. To that end, MutRoSe allows an MRS mission designer to define all relevant aspects of a mission and its environment in a high-level specification language that accounts for the variability of real-world scenarios, the dependencies between task instances, and the reusability of task libraries. Additionally, MutRoSe automates the decomposition of MRS missions defined in this language into task instances, which can then be allocated to specific robots for execution—with all task dependencies appropriately taken into account. We illustrate the application of MutRoSe and show its effectiveness for four missions taken from a recently published repository of MRS applications.

Evolving scheduling heuristics with genetic programming for optimization of quality of service in weakly hard real-time systems

The weakly hard real-time system model is used for describing the real-time systems that allow occasional violations of real-time timing constraints. These systems include real-time control systems, multimedia systems, and communication systems. In some approaches that deal with mitigating the system overload in real-time systems with periodic tasks, namely job-skipping algorithms, the constraints defined by the weakly hard real-time model are used as a mechanism for defining the pattern of task instances (jobs) that may be skipped in order to reduce the system load. The performance of these algorithms is usually evaluated with respect to the quality of service metric, which depends on the number of skipped jobs. In this work, we investigate the possibility of using genetic programming in the automated synthesis of scheduling heuristics for optimizing skipping patterns in order to increase the average quality of service in comparison with the conventional job-skipping algorithms. Using genetic programming to automatically synthesize heuristics allows for an easy and quick design of novel heuristics for various problem types and optimization criteria. We present two different approaches for implementing the proposed method. The first approach is to encapsulate the evolved heuristics into job-skipping algorithms known from the literature, namely Red Tasks as Late as Possible (RLP) and Blue When Possible (BWP). The idea of the second approach is to employ the evolved heuristics as standalone job-skipping algorithms. The results show an improvement of up to 15% in comparison with the state-of-the-art algorithms. The novel methods described in this work present a significant upgrade of the standard job-skipping algorithms as they provide a notable improvement in terms of quality of service while ensuring the fulfillment of weakly hard constraints. Moreover, the presented methods are computationally efficient and are therefore suitable for implementation on real-time operating systems, which is not the case with similar methods based on optimization techniques.

Changes in cortical activation during upright stance in individuals with chronic low back pain: An fNIRS study.

Postural control deficits are a potential cause of persistent and recurrent pain in patients with chronic low back pain (CLBP). Although some studies have confirmed that the dorsolateral prefrontal cortex (DLPFC) contributes to pain regulation in CLBP, its role in the postural control of patients with CLBP remains unclear. Therefore, this study aimed to investigate the DLPFC activation of patients with CLBP and healthy controls under different upright stance task conditions. Twenty patients with CLBP (26.50 ± 2.48 years) and 20 healthy controls (25.75 ± 3.57 years) performed upright stance tasks under three conditions: Task-1 was static balance with eyes open; Task-2 was static balance with eyes closed; Task-3 involved dynamic balance on an unstable surface with eyes open. A wireless functional near-infrared spectroscopy (fNIRS) system measured cortical activity, including the bilateral DLPFC, pre-motor cortex (PMC) and supplementary motor area (SMA), the primary motor cortex (M1), the primary somatosensory cortex (S1), and a force platform measured balance parameters during upright stance. The two-way repeated measures ANOVA results showed significant interaction in bilateral PMC/SMA activation. Moreover, patients with CLBP had significantly increased right DLPFC activation and higher sway 32 area and velocity than healthy controls during upright stance. Our results imply that PMC/SMA and DLPFC maintain standing balance. The patients with CLBP have higher cortical activity and upright stance control deficits, which may indicate that the patients with CLBP have low neural efficiency and need more motor resources to maintain balance.

Balance and Muscle Synergies During a Single-Limb Stance Task in Individuals With Chronic Ankle Instability.

The aim of this study was to investigate balance performance and muscle synergies during a Single-Limb Stance (SLS) task in individuals with Chronic Ankle Instability (CAI) and a group of healthy controls. Twenty individuals with CAI and twenty healthy controls were asked to perform a 30-second SLS task in Open-Eyes (OE) and Closed-Eyes (CE) conditions while standing on a force platform with the injured or the dominant limb, respectively. The activation of 13 muscles of the lower limb, hip, and back was recorded by means of surface electromyography. Balance performance was assessed by identifying the number and the duration of SLS epochs, and the Root-Mean-Square (RMS) in Antero-Posterior (AP) and Medio-Lateral (ML) directions of the body-weight normalized ground reaction forces. The optimal number of synergies, weight vectors, and activation coefficients were also analyzed. CAI group showed a higher number and a shorter duration of SLS epochs and augmented ground reaction force RMS in both AP and ML directions compared to controls. Both groups showed an increase in the RMS in AP and ML forces in CE compared to OE. Both groups showed 4 optimal synergies in CE, while controls showed 5 synergies in OE. CAI showed a significantly higher weight of knee flexor muscles in both OE and CE. In conclusion, muscle synergies analysis provided an in-depth knowledge of motor control mechanisms in CAI individuals. They showed worse balance performance, a lower number of muscle synergies in a CE condition and abnormal knee flexor muscle activation compared to healthy controls.

STRETCH: Virtual Shared-Nothing Parallelism for Scalable and Elastic Stream Processing

Stream processing applications extract value from raw data through Directed Acyclic Graphs of data analysis tasks. Shared-nothing (SN) parallelism is the de-facto standard to scale stream processing applications. Given an application, SN parallelism ins9tantiates several copies of each analysis task, making each instance responsible for a dedicated portion of the overall analysis, and relies on dedicated queues to exchange data among connected instances. On the one hand, SN parallelism can scale the execution of applications both up and out since threads can run task instances within and across processes/nodes. On the other hand, its lack of sharing can cause unnecessary overheads and hinder the scaling up when threads operate on data that could be jointly accessed in shared memory. This trade-off motivated us in studying a way for stream processing applications to leverage shared memory and boost the scale up (before the scale out) while adhering to the widely-adopted and SN-based APIs for stream processing applications. We introduce <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">STRETCH</i> , a framework that maximizes the scale up and offers instantaneous elastic reconfigurations (without state transfer) for stream processing applications. We propose the concept of Virtual Shared-Nothing (VSN) parallelism and elasticity and provide formal definitions and correctness proofs for the semantics of the analysis tasks supported by <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">STRETCH</i> , showing they extend the ones found in common Stream Processing Engines. We also provide a fully implemented prototype and show that <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">STRETCH</i> 's performance exceeds that of state-of-the-art frameworks such as Apache Flink and offers, to the best of our knowledge, unprecedented ultra-fast reconfigurations, taking less than 40 ms even when provisioning tens of new task instances.

Effect of Progressive Postural Control Exercise Versus Core Stability Exercise in Young Adults with Chronic Low Back Pain: A Randomized Controlled Trial.

This study aimed to investigate the effects of progressive postural control exercise (PPCE) vs core stability exercise (CSE) in patients with chronic low back pain (CLBP). A total of 34 young-adult participants with CLBP were randomly assigned to two groups (the PPCE group and the CSE group). They received instructions for two different exercise training regimens persisting over 8weeks. Before, after, and at 6months after the intervention, the participants were evaluated on the basis of pain intensity (VAS), degree of dysfunction (ODI and RMDQ), contractility of transversus abdominis (TrA) and lumbar multifidus (MF), as well as the ability to control static posture. There was no significant difference between the results of the PPCE group and the CSE group. At the 6-month follow-up after the 8-week treatment, the scores of VAS, ODI, and RMDQ in the two groups decreased significantly compared to before (p < 0.05). The percentage change in thickness of bilateral TrA and left MF (p < 0.05) was elevated and the sway area of center of pressure during static stance tasks with eyes opened (p < 0.05) was decreased in both groups. In the short term, PPCE provides positive effects similar to those of core stability exercise in patients with CLBP. The effective mechanism of PPCE might be the consequence of neuromuscular plasticity and adaptation adjustments. PPCE enriches the choices of treatment for CLBP. The trial was registered at www.chictr.org.cn , identifier ChiCTR2100043113.

Impact of Joint Fixation on Postural Dynamics during Single-Leg Stance

We investigated the relationship between the mechanical degrees of freedom (DoF) and its postural dynamics. The joint DoF was fixed to constrain the mechanical DoF. Nine participants were required to perform a single-leg stance task. The center of pressure trajectory data was analyzed. Ankle fixation induced a larger amount of variability in the anteroposterior direction, and less dimensionality and complexity in the mediolateral direction. These results suggest that the ankle joint fixation caused limited postural sway in the mediolateral direction; therefore, functional DoF and complexity decreased. In contrast, it increased the amount of postural sway variability in the anteroposterior direction. Our findings imply a direct relationship between the mechanical DoF of the human movement system and its postural dynamics.

An intelligent hybrid method: Multi-objective optimization for MEC-enabled devices of IoE

Emerging Internet-of-Everything (IoE) services require connected devices to respond instantly and operate for long durations. As smart mobile devices (SMDs) are often powered by batteries of limited capacity, offloading some computational tasks to nearby edge servers is a promising solution to reduce the latency and energy consumption of SMDs in operation. However, a challenge of computation offloading in the IoE system is the large number of SMDs that need to be handled. To address this problem, in this paper, we propose an intelligent two-stage computation offloading scheme with multiple optimization objectives. In the first stage, we categorize the computation tasks into classes (e.g., computation-intensive, data-intensive) and make early offloading decisions on some classes of tasks with offloading preferences. In the second stage, we make offloading decisions on the remaining tasks by solving a multi-objective optimization problem using the powerful Non-dominated Sorting Genetic Algorithm (NSGA-II). This two-stage design can help reduce the size of task instances in the second optimization stage, thus accelerating the convergence of the NSGA-II algorithm. The extensive simulation results show that, compared to the existing NSGA-II algorithm-based optimization methods, the proposed offloading scheme improves the performance by 10% in terms of latency and energy consumption.

Center of pressure characteristics differ during single leg stance throughout pregnancy and compared to nulligravida individuals

BackgroundFalls are common during pregnancy and present potential for injury to the pregnant individual and the baby. Research questionDo center of pressure characteristics during single leg stance differ between participants during and after pregnancy and nulligravida participants in the presence and absence of visual input? MethodsNineteen pregnant participants completed testing during the second trimester, the third trimester, and 4–6 months post-partum. Matched, nulligravida females completed testing once. All participants performed single leg stance on a force platform on each limb for up to 20 s with eyes open and with eyes closed. Center of pressure characteristics were compared between pregnant and nulligravida females using three separate 2 × 2 mixed way ANOVAs, one for each pregnancy time point (second trimester, third trimester, and post-partum) with Bonferroni correction. ResultsPregnant females demonstrated smaller single leg stance time with eyes closed during the third trimester. During the second and third trimester, pregnant participants demonstrated smaller sway and sway velocity across eyes open and eyes closed conditions. During the third trimester and post-partum, pregnant participants demonstrated greater median frequency of the center of pressure data. Pregnant participants also demonstrated smaller sample entropy in the anteroposterior direction during the second and third trimesters and in the mediolateral direction during the second trimester. SignificanceThe decreased total sway and sway velocity observed during pregnancy may reflect rigidity or a protective strategy during single limb stance. Additionally, center of pressure data were less smooth and more repetitive during pregnancy indicating robust differences in postural control strategies and potentially increased fall risk. Because single limb stance is a component of many activities of daily living, the single limb stance task may have clinical utility for testing or training balance in this population with a goal of decreasing falls.

System Comparison for Gait and Balance Monitoring Used for the Evaluation of a Home-Based Training.

There are currently no standard methods for evaluating gait and balance performance at home. Smartphones include acceleration sensors and may represent a promising and easily accessible tool for this purpose. We performed an interventional feasibility study and compared a smartphone-based approach with two standard gait analysis systems (force plate and motion capturing systems). Healthy adults (n = 25, 44.1 ± 18.4 years) completed two laboratory evaluations before and after a three-week gait and balance training at home. There was an excellent agreement between all systems for stride time and cadence during normal, tandem and backward gait, whereas correlations for gait velocity were lower. Balance variables of both standard systems were moderately intercorrelated across all stance tasks, but only few correlated with the corresponding smartphone measures. Significant differences over time were found for several force plate and mocap system-obtained gait variables of normal, backward and tandem gait. Changes in balance variables over time were more heterogeneous and not significant for any system. The smartphone seems to be a suitable method to measure cadence and stride time of different gait, but not balance, tasks in healthy adults. Additional optimizations in data evaluation and processing may further improve the agreement between the analysis systems.

RTOS-supported low power scheduling of periodic hardware tasks in flash-based FPGAs

As flash-based technology features inherently a lower power consumption than SRAM-based technology, flash-based FPGAs are well suited for energy-aware applications that are often found, e.g., in battery-powered embedded systems. Additionally, power gating allows to shut down large parts of an FPGA during idle times, reducing the energy consumption even further. When power gating SRAM-based FPGAs, the configuration memory is erased. The current state has to be stored and reconfigured in a time intense process before hardware accelerators can be used again. By contrast, in flash-based FPGAs the configuration memory and register contents are retained. In Microsemi / Microchip devices, the power gating mode is called Flash*Freeze. As only the complete FPGA can be put into Flash*Freeze mode, its applicability is reduced when many hardware tasks with different execution times share the same FPGA. This problem is tackled by the cluster scheduling algorithm presented in this paper. It modifies the schedule of hardware tasks under consideration of their real-time constraints in order to prolong Flash*Freeze phases. We describe its integration into the real-time operating system FreeRTOS, which allows to hide technical details from the applications and provides an easy-to-use interface. For a task set with sufficient overlapping task instances, the cluster scheduling algorithm reduces energy consumption on average by 33.5% in comparison to applications that do not employ the Flash*Freeze mode and by 22.5% in comparison to applications that use the Flash*Freeze mode with an unchanged schedule. In the best case, a prolongation of the Flash*Freeze mode in the order of n can be reached with n given hardware tasks.

Acute effect of inhibitory kinesio tape on range of motion, dynamic balance, and gait in athletes with hamstring shortness

ObjectiveThe goal of the present study was to investigate the acute effects of kinesio tape on range of motion (ROM), perceived stiffness, and kinetic parameters in athletes with hamstring shortness. MethodsFifteen athletes with bilateral hamstring shortness were divided randomly into intervention (inhibitory application of kinesio tape plus static stretching) and control (static stretching) groups. Outcome measures were straight leg raise (SLR), active knee extension (AKE) tests for ROM, visual analogue scale (VAS) for perceived stiffness and kinetic parameters which are recorded and calculated by Moticon SCIENCE insoles during four functional performance task of single leg stance (SLS), slow and fast walking and running. ResultsROM, stiffness, and kinetic parameters examined with insoles during SLS and walking tasks changed in both the intervention and control groups (P < 0.05). Furthermore, study groups were different in some variables. The increase in ROM in SLR test and the decrease in perceived stiffness in the intervention group were more evident than in the control group (P = 0.03 and P = 0.001, respectively). There was no difference between the intervention and control groups for the variables in SLS test, slow and fast walking, and running tasks, except for the mean swing duration (MSWD) (P = 0.002) in fast walking. ConclusionThe static stretching alone and the kinesio tape plus static stretching could change many of the kinetic parameters during walking and balance tasks, but the rate of these changes were more significant with kinesio tape plus stretching compared to the stretching alone.

Postural Sway in Parkinson's Disease and Multiple Sclerosis Patients During Tasks With Different Complexity.

Neurological diseases are associated with static postural instability. Differences in postural sway between neurological diseases could include “conceptual” information about how certain symptoms affect static postural stability. This information might have the potential to become a helpful aid during the process of finding the most appropriate treatment and training program. Therefore, this study investigated static postural sway performance of Parkinson's disease (PD) and multiple sclerosis (MS) patients, as well as of a cohort of healthy adults. Three increasingly difficult static postural tasks were performed, in order to determine whether the postural strategies of the two disease groups differ in response to the increased complexity of the balance task. Participants had to perform three stance tasks (side-by-side, semi-tandem and tandem stance) and maintain these positions for 10 s. Seven static sway parameters were extracted from an inertial measurement unit that participants wore on the lower back. Data of 47 healthy adults, 14 PD patients and 8 MS patients were analyzed. Both healthy adults and MS patients showed a substantial increase in several static sway parameters with increasingly complex stance tasks, whereas PD patients did not. In the MS patients, the observed substantial change was driven by large increases from semi-tandem and tandem stance. This study revealed differences in static sway adaptations between PD and MS patients to increasingly complex stance tasks. Therefore, PD and MS patients might require different training programs to improve their static postural stability. Moreover, this study indicates, at least indirectly, that rigidity/bradykinesia and spasticity lead to different adaptive processes in static sway.

Effects of Fatigue on Postural Sway and Electromyography Modulation in Young Expert Acrobatic Gymnasts and Healthy Non-trained Controls During Unipedal Stance.

This study investigated whether expert acrobatic gymnasts respond differentially than their non-trained counterparts during a single-legged stance task performed before and after a protocol designed to induce fatigue in the ankle plantarflexor muscles in terms of (a) postural steadiness and (b) electromyography (EMG) activation. We hypothesized that neuromuscular adaptation due to training would lead to different behavior of center of pressure (COP) and EMG quantifiers after fatigue. Twenty eight female volunteers (aged 11 to 24 years) formed two groups: expert acrobatic gymnastics athletes (GYN, n = 14) and age-matched non-gymnasts [control (CTRL), n = 14]. Fatigue of the ankle plantarflexors (dominant leg) was induced by a sustained posture (standing on the toes) until exhaustion. Traditional COP parameters (area, RMS, mean velocity, and power spectrum at low and high frequency ranges) were obtained with a force plate, and time and frequency-domain EMG parameters were obtained by surface electrodes positioned on the tibialis anterior, soleus, lateral gastrocnemius, medial gastrocnemius, vastus lateralis, biceps femoris, spinal erector and rectus abdominis muscles. The main results showed that fatigue induced a significant increase in postural oscillations in the ML axis (including RMS, velocity and frequency components of the power spectrum), with no significant effects in the AP axis. In terms of postural sway parameters (i.e., COP quantifiers), no superior balance stability was found for the GYN group as compared to CTRL, irrespective of the fatigue condition. On the other hand, the modulation of EMG parameters (in both time and frequency domains) indicated that expert acrobatic gymnastics athletes (as compared to healthy untrained matched controls) used different neuromuscular control strategies to keep their postures on single-legged quiet standing after the fatiguing protocol. The present results improve our knowledge of the mechanisms behind the interplay between fatigue and postural performance associated with the neuromuscular adaptations induced by sport practice. The design of gymnastics training might consider strategies aimed at improving the performance of specific muscles (i.e., tibialis anterior, soleus, biceps femoris, spinal erector) for which particular activation patterns were used by the acrobatic gymnastics to control single-legged quiet standing.

Workflow performance prediction based on graph structure aware deep attention neural network

• In workflow performance prediction, DAG structure matters; • DAG-Transformer effectively embeds the DAG information and outperforms mainstream ML, DL , and GCN methods; • A new dataset for cloud workflow performance prediction is accompanied as well as the source code. With the rapid growth of cloud computing, efficient operational optimization and resource scheduling of complex cloud business processes rely on real-time and accurate performance prediction. Previous research on cloud computing performance prediction focused on qualitative (heuristic rules), model-driven, or coarse-grained time-series prediction, which ignore the study of historical performance, resource allocation status and service sequence relationships of workflow services. There are even fewer studies on prediction for workflow graph data due to the lack of available public datasets. In this study, from Alibaba Cloud's Cluster-trace-v2018, we extract nearly one billion offline task instance records into a new dataset, which contains approximately one million workflows and their corresponding directed acyclic graph (DAG) matrices. We propose a novel workflow performance prediction model (DAG-Transformer) to address the aforementioned challenges. In DAG-Transformer, we design a customized position encoding matrix and an attention mask for workflows, which can make full use of workflow sequential and graph relations to improve the embedding representation and perception ability of the deep neural network. The experiments validate the necessity of integrating graph-structure information in workflow prediction. Compared with mainstream deep learning (DL) methods and several classic machine learning (ML) algorithms, the accuracy of DAG-Transformer is the highest. DAG-Transformer can achieve 85-92% CPU prediction accuracy and 94-98% memory prediction accuracy, while maintaining high efficiency and low overheads. This study establishes a new paradigm and baseline for workflow performance prediction and provides a new way for facilitating workflow scheduling.

An Adaptive Imitation Learning Framework for Robotic Complex Contact-Rich Insertion Tasks.

Complex contact-rich insertion is a ubiquitous robotic manipulation skill and usually involves nonlinear and low-clearance insertion trajectories as well as varying force requirements. A hybrid trajectory and force learning framework can be utilized to generate high-quality trajectories by imitation learning and find suitable force control policies efficiently by reinforcement learning. However, with the mentioned approach, many human demonstrations are necessary to learn several tasks even when those tasks require topologically similar trajectories. Therefore, to reduce human repetitive teaching efforts for new tasks, we present an adaptive imitation framework for robot manipulation. The main contribution of this work is the development of a framework that introduces dynamic movement primitives into a hybrid trajectory and force learning framework to learn a specific class of complex contact-rich insertion tasks based on the trajectory profile of a single task instance belonging to the task class. Through experimental evaluations, we validate that the proposed framework is sample efficient, safer, and generalizes better at learning complex contact-rich insertion tasks on both simulation environments and on real hardware.