Multi-agent reinforcement learning (MARL) suffers from the non-stationarity problem, which is ever-changing targets at every iteration when multiple agents update their policies same time. Starting first principle, in this paper, we manage to solve problem by proposing bidirectional action-dependent Q-learning (ACE). Central development of ACE sequential decision making process wherein only one agent allowed take action Within process, each maximizes its value function given actions taken...

Numerous tissue-engineered constructs have been investigated as bone scaffolds in regenerative medicine. However, it remains challenging to non-invasively monitor the biodegradation and remodeling of grafts after implantation. Herein, silk fibroin/hydroxyapatite incorporated with ultrasmall superparamagnetic iron oxide (USPIO) nanoparticles were successfully synthesized, characterized, implanted subcutaneously into back nude mice. The USPIO labeled showed good three-dimensional porous...

Current constrained reinforcement learning (RL) methods guarantee constraint satisfaction only in expectation, which is inadequate for safety-critical decision problems. Since a satisfied expectation remains high probability of exceeding the cost threshold, solving RL problems with probabilities critical safety. In this work, we consider safety criterion as on conditional value-at-risk (CVaR) cumulative costs, and propose CVaR-constrained policy optimization algorithm (CVaR-CPO) to maximize...

The Cartesian path planning of free-floating space robot is much more complex than that fixed-based manipulators, since the end-effector pose (position and orientation) dependent, position-level kinematic equations can not be used to determine joint angles. In this paper, a method based on particle swarm optimization (PSO) proposed solve problem. Firstly, we parameterize trajectory using polynomial functions, then normalize parameterized trajectory. Secondly, transformed an problem by...

Efficient and effective exploration in continuous space is a central problem applying reinforcement learning (RL) to autonomous driving. Skills learned from expert demonstrations or designed for specific tasks can benefit the exploration, but they are usually costly-collected, unbalanced/suboptimal, failing transfer diverse tasks. However, human drivers adapt varied driving without by taking efficient structural explorations entire skill rather than limited with task-specific skills....

In this paper, we focus on the integration of motion planning and control for high-performance automated vehicles. We propose a novel tube-based nonlinear model predictive (TNMPC) scheme that combines strengths MPC boundary layer sliding (BLSC) to drive controlled vehicle along real-time trajectory while simultaneously ensuring practical satisfaction constraints. Specifically, our approach uses an extended kinematic with slip angle parameters provide explicit expressions modeling error,...

Four-wheel, independently driven skid-steer mobile robots have been widely used in some fields, such as indoor product shipping and outdoor inspection exploration. Traditional robot controllers often use a kinematics controller to obtain the desired speed of each wheel, complete closed-loop control wheel achieve robot’s trajectory tracking control. However, based on may lead chattering spin from being directly by motor uneven grounds. To solve these problems, we developed four-wheel, with...

Deep learning (DL) has made tremendous contributions to image processing. Recently, the DL also attracted attention in specialized field of neural decoding from raw myoelectric signals (electromyograms, EMGs). However, our knowledge, most existing methods require some measure preprocessing EMGs. Moreover, research date not accounted for variability signal during time sequences. In this paper, we propose a new convolutional network (CNN) structure that can directly process EMG hand gesture...

Cable-Constrained Synchronous Rotating Mechanism (CCSRM) has an important application prospect in the field of cable-driven robots, which can greatly reduce number driving motors while ensuring light and slender body. However, there are obvious cable friction effect elastic deformation CCSRM. These nonlinear characteristics have a significant impact on synchronous motion performance. In this letter, model CCSRM considering is proposed, integrates effects pretension, deformation, between...

In this work we introduce reinforcement learning techniques for solving lexicographic multi-objective problems. These are problems that involve multiple reward signals, and where the goal is to learn a policy maximises first signal, subject constraint also second so on. We present family of both action-value gradient algorithms can be used solve such problems, prove they converge policies lexicographically optimal. evaluate scalability performance these empirically, demonstrating their...

Benefiting from the powerful learning capacity of deep (DP), a general model for predicting 3-DOF wrist movements could achieve good prediction accuracy those subjects even not involved in training. This paper tends to verify this assumption. Since quantity training dataset DP largely influences model's performance, limited collected number should be extended first through some data-augmentation approaches. In paper, we summarized possible mistakes happen EMG data-collection procedures....

The reaction wheel bicycle robot is a kind of unmanned mobile with great potential. However, the control such robots on curved pavement under inaccurate model parameters, uncertainties and disturbances challenging due to lateral instability udneractuated characteristic. Applying conventional methods this problem often results in brittle controllers. In paper, an online serial-parallel combination reinforcement learning designed achieve path tracking banlancing for pavements. parallel part...

Abstract Recent studies have found that deep learning models are vulnerable to adversarial examples, demonstrating applying a certain imperceptible perturbation on clean examples can effectively deceive the well-trained and high-accuracy models. Moreover, achieve considerable level of certainty with attacked label. In contrast, human could barely discern difference between which raised tremendous concern about robust trustworthy techniques. this survey, we reviewed existence, generation,...

Fast response heat flux gauges relying on the semi-infinite conduction principle have commonly been used to study in impulse test facilities such as expansion tubes and reflected shock tunnels. For studying very harsh environments experienced at stagnation point of entry vehicles, generally metallic thermocouples are required survive loads which may be above ten megawatts per metre squared peak heating flight can upwards a hundred heavily scaled facility testing, challenging for even...

The application requirements of the tendon-sheath mechanism in field precision machinery are becoming increasingly extensive. However, contact friction between tendon and sheath seriously affects transmission accuracy. In case unavoidable friction, optimizing path to reduce tension loss elastic deformation has become an important research direction. this article, influence law on displacement is obtained using two parameters related curvature path: total bending angle equivalent length....

The paper presents a new calibration method of the parallel robot using genetic algorithm (GA) which has powerful global adaptive probabilistic search ability. In process calibration, calculation measurement residuals through inverse kinematics is adopted, greatly reduces burden and saves operation time. At last, in form simulation experiment, considering influence noise steward respectively calibrated its pose (position orientation) accuracy significant improvement.