Accelerated Reduced Gradient Algorithm with Constraint Relaxation in Differential Inverse Kinematics

The Moore-Penrose pseudoinverse-based solution of the differential inverse kinematic task of redundant robots corresponds to the result of a particular optimization underconstraints in which the implementation of Lagrange’s ReducedGradient Algorithm can be evaded simply by considering the zero partial derivatives of the ”Auxiliary Function” associated with this problem. This possibility arises because of the fact that the cost term is built up of quadratic functions of the variable of optimization while the constraint term is linear function of the same variables. Any modification in the cost and/or constraint structure makes it necessary the use of the numerical algorithm. Anyway, the penalty effect of the cost terms is always overridden by the hard constraints that makes practical problems in the vicinity of kinematic singularities where the possible solution stillexists but needs huge joint coordinate time-derivatives. While in the special case the pseudoinverse simply can be deformed, inthe more general one more sophisticated constraint relaxation can be applied. In this paper a formerly proposed acceleratedtreatment of the constraint terms is further developed by the introduction of a simple constraint relaxation. Furthermore, thenumerical results of the algorithm are smoothed by a third order tracking strategy to obtain dynamically implementable solution.The improved method’s operation is exemplified by computation results for a 7 degree of freedom open kinematic chain

Timetabling problem solving based on best-nests cuckoo search

The university courses timetabling problem (UCTP) is a popular subject among institutions and academics because occurs every academic year. In general, UCTP is the distribution of events through slots time for each room based on the list of constraints for instance (hard constraint and soft constraint) supplied in one semester, intending to avoid conflicts in such assignments. Under no circumstances should hard constraints be broken while attempting to fulfill as many soft constraints as feasible. this article presented a modified best-nests cuckoo search (BNCS) algorithm depend on the base cuckoo search (CS) algorithm. BNSC algorithm was achieved by dividing the nests into two groups (best-nests and normal-nests). The BNCS algorithm selection was limited to the best-nests to generate new solutions. The comparison between BNCS and basic CS based on the experimental result is achieved. For performance evaluation, the BNCS has been tested on four variant-size datasets. It was observed that the BNCS has performed high performance and is faster at finding a solution from CS.

Parallel Hybrid Electric Vehicle Modelling and Model Predictive Control

This paper presents the modelling and calculations for a hybrid electric vehicle (HEV) in parallel configuration, including a main electrical driving motor (EM), an internal combustion engine (ICE), and a starter/generator motor. The modelling equations of the HEV include vehicle acceleration and jerk, so that simulations can investigate the vehicle drivability and comfortability with different control parameters. A model predictive control (MPC) scheme with softened constraints for this HEV is developed. The new MPC with softened constraints shows its superiority over the MPC with hard constraints as it provides a faster setpoint tracking and smoother clutch engagement. The conversion of some hard constraints into softened constraints can improve the MPC stability and robustness. The MPC with softened constraints can maintain the system stability, while the MPC with hard constraints becomes unstable if some input constraints lead to the violation of output constraints.

Multi-Label Classification Neural Networks with Hard Logical Constraints

Multi-label classification (MC) is a standard machine learning problem in which a data point can be associated with a set of classes. A more challenging scenario is given by hierarchical multi-label classification (HMC) problems, in which every prediction must satisfy a given set of hard constraints expressing subclass relationships between classes. In this article, we propose C-HMCNN(h), a novel approach for solving HMC problems, which, given a network h for the underlying MC problem, exploits the hierarchy information in order to produce predictions coherent with the constraints and to improve performance. Furthermore, we extend the logic used to express HMC constraints in order to be able to specify more complex relations among the classes and propose a new model CCN(h), which extends C-HMCNN(h) and is again able to satisfy and exploit the constraints to improve performance. We conduct an extensive experimental analysis showing the superior performance of both C-HMCNN(h) and CCN(h) when compared to state-of-the-art models in both the HMC and the general MC setting with hard logical constraints.

Emotion-based Parameter Modulation for a Mobile Robot Planning and Control System

<p>The hypothesis that artificial emotion-like mechanisms can improve the adaptive performance of robots and intelligent systems has gained considerable support in recent years. While artificial emotions are typically employed to facilitate human-machine interaction, this thesis instead focuses on modelling emotions and affect in a non-social context. In particular, affective mechanisms are applied to the problem of mobile robot navigation. A three-layered reactive/deliberative controller is developed and implemented, resulting in several contributions to the field of mobile robot control. Rather than employing a reactive layer, a deliberative layer and an interface between them, the control problem is decomposed into three different conceptual spaces - position space, direction space and velocity space - with a distinct control layer applied to each. Existing directional and velocity space approaches such as the vector field histogram (VFH) and dynamic window methods employ different underlying mechanisms and terminology. This thesis unifies these approaches in order to compare and combine them. The weighted sum objective functions employed by some existing approaches that inspired the presented directional and velocity control layers are replaced by weighted products. This enables some hard constraints to be relaxed in favour of weighted contributions, potentially improving a system's flexibility without sacrificing safety (but coming at a cost to efficiency). An affect model is developed that conceptualises emotions and other affective interactions as modulations of cognitive processes. Unlike other models of affect-modulated cognition (e.g. Dorner and Hille, 1995), this model is designed specifically to address problems relating to mobile robot navigation. The role of affect in this model is to continuously adapt a controller's behaviour patterns in response to different environments and momentary conditions encountered by the robot. Affective constructs such as moods and emotions are represented as intensity values that arise from hard-coded interpretations of local stimuli, as well as from learned associations stored in global maps. They are expressed as modulations of control parameters and location-specific biases to path-planning. Extensive simulation experiments are conducted in procedurally-generated environments to assess the performance contributions of this model and its individual components.</p>

Emotion-based Parameter Modulation for a Mobile Robot Planning and Control System

<p>The hypothesis that artificial emotion-like mechanisms can improve the adaptive performance of robots and intelligent systems has gained considerable support in recent years. While artificial emotions are typically employed to facilitate human-machine interaction, this thesis instead focuses on modelling emotions and affect in a non-social context. In particular, affective mechanisms are applied to the problem of mobile robot navigation. A three-layered reactive/deliberative controller is developed and implemented, resulting in several contributions to the field of mobile robot control. Rather than employing a reactive layer, a deliberative layer and an interface between them, the control problem is decomposed into three different conceptual spaces - position space, direction space and velocity space - with a distinct control layer applied to each. Existing directional and velocity space approaches such as the vector field histogram (VFH) and dynamic window methods employ different underlying mechanisms and terminology. This thesis unifies these approaches in order to compare and combine them. The weighted sum objective functions employed by some existing approaches that inspired the presented directional and velocity control layers are replaced by weighted products. This enables some hard constraints to be relaxed in favour of weighted contributions, potentially improving a system's flexibility without sacrificing safety (but coming at a cost to efficiency). An affect model is developed that conceptualises emotions and other affective interactions as modulations of cognitive processes. Unlike other models of affect-modulated cognition (e.g. Dorner and Hille, 1995), this model is designed specifically to address problems relating to mobile robot navigation. The role of affect in this model is to continuously adapt a controller's behaviour patterns in response to different environments and momentary conditions encountered by the robot. Affective constructs such as moods and emotions are represented as intensity values that arise from hard-coded interpretations of local stimuli, as well as from learned associations stored in global maps. They are expressed as modulations of control parameters and location-specific biases to path-planning. Extensive simulation experiments are conducted in procedurally-generated environments to assess the performance contributions of this model and its individual components.</p>

Optimizing the Tolerance for the Products with Multi-Dimensional Chains via Simulated Annealing

The assembly is the last process of controlling the product quality during manufacturing. The installation guidance should provide the appropriate assembly information, e.g., to specify the components in each product. The installation guidance with low quality results in rework or the resource waste from the failure products. This article extends the dimensional chain assembly problem proposed by Tsung et al. to consider the multiple dimensional chains in the product. Since there are multiple dimensional chains in a product, the installation guidance should consider inseparability and acceptability as computing the installation guidance. The inseparability means that the qualities of all dimensional chains in the part should be evaluated together without separation, while the acceptability stands for that the size of each product should be satisfied with the specification. The simulated annealing (SA) algorithm is applied to design the assembly guidance optimizer named as AGOMDC to compute the assembly guidance in the dimensional chain assembly problem with multiple dimensional chains. Since SA has high performance in searching neighbor solutions, the proposed approach could converge rapidly. Thus, proposed AGOMDC could be applied in real-world application for the implementation consideration. The simulations consist of two parts: the feasibility evaluation and the algorithm configuration discussion. The first part is to verify the inseparability and acceptability that are the hard constraints of the assembly problem for the proposed AGOMDC, and the second one is to analyze the algorithm configurations to calculate the assembly guidance with 80% quality. The simulation results show that the inseparability and acceptability are achieved, while the proposed AGOMDC only requires more than two seconds to derive the results. Moreover, the recommended algorithm configurations are derived for evaluate the required running time and product quality. The configurations with product quality 80% are that the temperature descent rate is 0.9, the initial temperature is larger than 1000, and the iteration recommended function is derived based on the problem scale. The proposed AGOMDC not only helps the company to save the time of rework and prevent the resource waste of the failure products, but is also valuable for the automatic assembly in scheduling the assembly processes.

USING A GENETIC ALGORITHM TO SOLVE THE COURSES TIMETABLING CREATION PROBLEM

Creating of courses timetable is an extremely difficult, time-consuming task and usually takes a long time. In many educational institutions, the courses schedule is developed manually. Schedule theory includes problems that are actually less complex than problems in practice, but theoretical analysis provides a fundamental understanding of the complexity of the schedule. The logical result is that the schedule is very difficult to build in practice due to many constraints [1]. Scheduling courses is a planning problem. In 1996, the problem of scheduling was described as the allocation of some resources with restrictions on a limited number of time intervals and at the same time to satisfy the set of stated objectives [2]. This is a general statement and is a common description of the courses timetabling creation problem. Schedule of courses is an important administrative activity in most educational institutions. The timetable problem is the distribution of classes by available audiences and time intervals, taking into account the constraints. We usually distinguish between two types of constraints: hard and soft. Hard constraints are compulsorily fulfilled by the educational institution. Decisions that do not violate hard constraints are called possible solutions. With the development of the general theory of the schedule, the approaches to the formalization and solution of the courses timetabling creation problem in educational institutions also changed. Currently, the problem of automation of the courses timetabling creation remains relevant. The urgency of the problem is determined by the growing requirements for the quality of education, student work planning, rational use of the audiences, as well as taking into account additional optimization parameters. The task of finding the optimal schedule of courses in most cases belongs to the class of complex problems. If we take into account the real conditions, the problem is even more complicated, because the desired solutions must meet numerous constraints of production, organizational and psychophysiological nature, which contradict each other. The genetic algorithm helps to efficiently search for optimal solutions in spaces with a very large dimension.

Learning without loss

We explore a new approach for training neural networks where all loss functions are replaced by hard constraints. The same approach is very successful in phase retrieval, where signals are reconstructed from magnitude constraints and general characteristics (sparsity, support, etc.). Instead of taking gradient steps, the optimizer in the constraint based approach, called relaxed–reflect–reflect (RRR), derives its steps from projections to local constraints. In neural networks one such projection makes the minimal modification to the inputs x, the associated weights w, and the pre-activation value y at each neuron, to satisfy the equation $x\cdot w=y$ x ⋅ w = y . These projections, along with a host of other local projections (constraining pre- and post-activations, etc.) can be partitioned into two sets such that all the projections in each set can be applied concurrently—across the network and across all data in the training batch. This partitioning into two sets is analogous to the situation in phase retrieval and the setting for which the general purpose RRR optimizer was designed. Owing to the novelty of the method, this paper also serves as a self-contained tutorial. Starting with a single-layer network that performs nonnegative matrix factorization, and concluding with a generative model comprising an autoencoder and classifier, all applications and their implementations by projections are described in complete detail. Although the new approach has the potential to extend the scope of neural networks (e.g. by defining activation not through functions but constraint sets), most of the featured models are standard to allow comparison with stochastic gradient descent.