Although robotic trajectory generation problem has been extensively investigated, existing solutions are almost customized to specific robot geometry, and generalized schemes are yet to be explored. In this article, a general motion planning framework based on deep reinforcement learning (DRL) and artificial neural networks (ANNs) is proposed for robot with arbitrary geometry. First, a unique screening and grafting mechanism is established to improve the policy learning by exploiting valuable experience sufficiently. Second, based on the reward-oriented characteristics of DRL, a forward progression mechanism is proposed to facilitate the path planning for complicated tasks. Third, a structure consisting of an adventurer and conservator algorithm with automatic optimization and an ANN-based mapper is designed integrally to derive the inverse kinematics solutions without considering the robot geometry. Finally, experimental results have verified the superior performance of the proposed approach.Note to Practitioners—This article is aiming to provide a general method to solve the problem of motion planning for robots via deep reinforcement learning (DRL) and artificial neural networks (ANNs). Compared to the existing approaches, which are highly specialized and limited to robots with specific geometries, and often cumbersome, our method can be easily applied to robots with arbitrary geometries and has good generalization, where to simplify the training based on DRL for diverse practical motion planning tasks, a universal forward progression mechanism is used to partition a complex task into multiple successive simple phases. Furthermore, an ANN-based mapper can cope with the burdensome inverse kinematics of robots with both common and uncommon geometries, especially the ones with redundant degrees of freedom. The proposed method is also validated to be superior to other state-of-the-art methods in real-world experimental studies.

The information transformation problem of model driven software development (MDSD) processes is gaining increasing attentions. In this paper, we present an indepth analysis of the major causes leading to the ambiguities of existing transformation. A metamodel based approach is then proposed to address these causes. A mechanism to generate transformation rules is devised based on a classification of the UML metamodel elements to be used in subsequent transformation processes. Some other critical issues relating to the generated transformation system are also discussed.

This paper investigates the integration of robotic task programming and assembly sequence planning in virtual assembly. It presents an integrated approach that derives assembly relationship of machinery parts as required by sequence planning automatically from the information provided by interactive programming. 'Automatic reprogramming' also becomes accessible, which feeds alternative assembly sequences back to programming for optimization.