Multi-objective grasp pose optimisation for robotic 3D pipe assembly manipulation

Abstract

This paper considers the problem of grasp pose optimisation for manipulating 3D pipe assemblies during the manufacturing process. The method presented in this paper is specifically developed for manufacturing cryogenic pipe assemblies autonomously in a Factory-In-A-Box scenario (i.e., a compact factory built inside an industrial container). However, it also can be used for robotic manipulation of general frame structures. The problem is formulated as a constrained multi-objective optimisation problem. The optimisation algorithm searches for solutions that satisfy two constraints: (i) robot workspace reachability (i.e., feasible inverse kinematics solution for a given end-effector pose); and (ii) any possible collision when executing the post-grasp trajectory, and continuously improves them based on three objectives: (i) minimise robot joint motion for executing a specified pipe trajectory; (ii) minimise the sum of maximum deformation of pipe assembly along the trajectory; and (iii) minimise the sum of robot force manipulability along the trajectory. A special constraint handling method is used to decouple the constraint and objective evaluation process, allowing expensive objectives to be evaluated only when constraints are satisfied. The algorithm explicitly considers the possibility of multiple inverse kinematics solutions and uses a graph search algorithm (Dijkstra's Algorithm) to find the optimal trajectory amongst all feasible trajectories. The weighted sum approach is used to combine the three objectives with weights determined by the Analytical Hierarchy Process. The optimisation problem is solved using the Bees Algorithm with a proposed problem-specific local search strategy. Extensive benchmarks show that the proposed strategy achieves better overall results than the default strategy of the Bees Algorithm and other metaheuristics.