Trajectory optimization for an automated stock-picking robot: A review and experimental evaluation

Abstract: To address the issues related to the physically demanding task of stock-picking that exists in today’s logistics industry, robots have been suggested as a means to reduce the human workload. Extensive efforts have been made by the robotics community to solve problems related to automatic stock picking. Among these problems is motion planning, which is one of the most active fields of robotics. Motion planning for robot manipulator is a complex problem due to the high dimensionality of the problem. To address this problem, trajectory optimization is introduced as an approach to solve motion planning problems in higher dimensions. This thesis reviews and evaluates trajectory optimization methods for a stock-picking robot developed at KTH. The thesis is separated into two parts, which consists of a literature study and experimentation. In the literature study, we discuss previous work within trajectory optimization and the various strengths and weaknesses. The experiments are conducted on the trajectory optimization methods GPMP2, CHOMP and STOMP in different pick-and-place scenarios. The goal of these experiments is to evaluate their performance, the quality of the trajectories they generate and their applicability for pick-and-place tasks with KTH:s ASP robot. By evaluating the results from our experiments, we conclude that the evaluated trajectory optimization methods achieved limited success in the pick-and-place scenarios. The overall performance of STOMP meets the criteria, but has difficulties in narrow passages. The experiments also showed that the addition of a kernel smoother in the obstacle cost function improve the performance of STOMP.