Page - 50 - in Joint Austrian Computer Vision and Robotics Workshop 2020

AutomatedLogOrdering throughRoboticGrasper StephanWeiss1, StefanAinetter2, FredArneitz1,DailysArrondePerez1,RohitDhakate1, FriedrichFraundorfer2,HaraldGietler1,WolfgangGubensa¨k3, MylenaMedeirosDosReisFerreira1,ChristianStetco1,HubertZangl1 1)UniversityofKlagenfurt {[first given name].[first surname]}@aau.at 2)GrazUniversityofTechnology {stefan.ainetter, fraundorfer}@icg.tugraz.at 3)SpringerMaschinenfabrikGmbH Wolfgang.Gubensaek@springer.eu ∗ Abstract. This work focuses on retrofitting a crane model in the wood industry for automated log grasping. AI inspired vision based approaches are used to categorize and segment the logs and their geometry to subsequently define optimal grasping poses. Retrofittable sensors and robust control strategies for cost efficient upgrading of existing manually operated cranes towards autonomous systemsare developed. 1. Introduction Classical production lines and handling processes for raw materials often have a long history and incorporate a large amount of experience based knowledge for process optimization and handling routines. Nowadays, these processes seem to be stuck in a local minima in terms of efficiency and performance due to human factors. With the available degree of automation, robustness of AI based perception and decision making, and novel sensor technology, a re-thinking of these well established processes can take place. Instead of a radical approach to replace existing infrastructure, thiswork leveragescurrently installed machines in the wood sector and enables them to work autonomously through retro-fitting of sensors, autonomy, and AI based scene understanding. The project has a strong focus on bringing advanced methods in the corresponding research fields to practice. Hence, a model log crane was built as a 1:5 scaled downcopyofa real logcrane (Fig.1). ∗ The research leading to these resultshas received funding fromBMVITundergrantn. 864807 (AutoLOG) Figure1.Cranemodel in1:5scaledversionofarealcrane used in the wood sector. The hydraulics are specifically designed to match this scale. Manual control is identical to the realversions. 2.ModelandRetrofittableSensorDesign The1:5scaledcranemodelhasbeendesignedand manufactured from scratch to match the properties of the real counterparts. This includes hydraulic actuators, end-effector with two free joints and an actuated revolute joint with unconstrained 360◦ actuation, andbacklash. For tests andevaluation,we installed wire-rope sensors on the hydraulic pistons to measure their current position. Novel capacitive and inductive sensors have been designed and implemented as described in Section 2.1 to measure the current absolute angles and to provide feedback on the grasping quality. Apart of the crane itself, the overall system (Fig. 2) also contains a log storage boxwithautomatedemptyingmechanism. Emptying 50