The past decade has witnessed tremendous progress in soft robotics. Unlike most pneumatic-based methods, we present a new approach to soft robot design based on precharged pneumatics (PCP). We propose a PCP soft bending actuator, which is actuated by precharged air pressure and retracted by inextensible tendons. By pulling or releasing the tendons, the air pressure in the soft actuator is modulated, and hence, its bending angle. The tendons serve in a way similar to pressure-regulating valves that are used in typical pneumatic systems. The linear motion of tendons is transduced into complex motion via the prepressurized bent soft actuator. Furthermore, since a PCP actuator does not need any gas supply, complicated pneumatic control systems used in traditional soft robotics are eliminated. This facilitates the development of compact untethered autonomous soft robots for various applications. Both theoretical modeling and experimental validation have been conducted on a sample PCP soft actuator design. A fully untethered autonomous quadrupedal soft robot and a soft gripper have been developed to demonstrate the superiority of the proposed approach over traditional pneumatic-driven soft robots.

No abstract

Flexible flow shop scheduling problems are NP-hard and tend to become more complex when stochastic uncertainties are taken into consideration. Although some methods have been developed to address such problems, they remain inherently difficult to solve by any single approach. This paper presents a novel decomposition-based approach (DBA), which combines both the Shortest Processing Time (SPT) and the Genetic Algorithm (GA), to minimizing the makespan of a flexible flow shop (FFS) with stochastic processing times. In the proposed DBA, a neighbouring K-means clustering algorithm is developed to firstly group the machines of an FFS into an appropriate number of machine clusters, based on their stochastic nature. Two optimal back propagation networks (BPN), corresponding to the scenarios of simultaneous and non-simultaneous job arrivals, are then selectively adopted to assign either SPT or GA to each machine cluster for sub-schedule generation. Finally, an overall schedule is generated by integrating the sub-schedules of machine clusters.Computation results show that the DBA outperforms SPT and GA alone for FFS scheduling with stochastic processing times.

No abstract

The permutation flowshop scheduling problem (PFSP) is NP-complete and tends to be more complicated when considering stochastic uncertainties in the real-world manufacturing environments. In this paper, a two-stage simulation-based hybrid estimation of distribution algorithm (TSSB-HEDA) is presented to schedule the permutation flowshop under stochastic processing times. To deal with processing time uncertainty, TSSB-HEDA evaluates candidate solutions using a novel two-stage simulation model (TSSM). This model first adopts the regression-based meta-modelling technique to determine a number of promising candidate solutions with less computation cost, and then uses a more accurate but time-consuming simulator to evaluate the performance of these selected ones. In addition, to avoid getting trapped into premature convergence, TSSB-HEDA employs both the probabilistic model of EDA and genetic operators of genetic algorithm (GA) to generate the offspring individuals. Enlightened by the weight training process of neural networks, a self-adaptive learning mechanism (SALM) is employed to dynamically adjust the ratio of offspring individuals generated by the probabilistic model. Computational experiments on Taillard's benchmarks show that TSSB-HEDA is competitive in terms of both solution quality and computational performance.

No abstract

In this paper we apply a k-nearest-neighbor-based data condensing algorithm to the training set of multilayer perceptron neural networks. By removing the overlapping data and retaining only training exemplars adjacent to the decision boundary we are able to significantly speed the network training time while achieving an undegraded misclassification rate compared to a network trained on the unedited training set. We report results on a range of synthetic and real datasets that indicate that a training speed-up of an order of magnitude is typical.

No abstract