Journal of Manufacturing Systems 32 (2013) 429–435 Contents lists available at SciVerse ScienceDirect Journal of Manufacturing Systems jo ur nal home p age: www.elsevier.com/locate/jmansys Non-nominal path planning for robust robotic assembly Johan S. Carlson a , Domenico Spensieri a,∗ , Rikard Söderberg b , Robert Bohlin a , Lars Lindkvist b a Fraunhofer-Chalmers Centre, Johanneberg Science Park, Göteborg, Sweden b Wingquist Laboratory, Chalmers University of Technology, Göteborg, Sweden a r t i c l e i n f o Article history: Received 15 September 2012 Received in revised form 23 April 2013 Accepted 26 April 2013 Available online 25 May 2013 Keywords: Assembly Computer aided manufacturing Dimensional control Path planning Robotics Quality assurance a b s t r a c t In manufacturing and assembly processes it is important, in terms of time and money, to verify the feasi- bility of the operations at the design stage and at early production planning. To achieve that, verification in a virtual environment is often performed by using methods such as path planning and simulation of dimensional variation. Lately, these areas have gained interest both in industry and academia, however, they are almost always treated as separate activities, leading to unnecessary tight tolerances and on-line adjustments. To resolve this, we present a novel procedure based on the interaction between path planning tech- niques and variation simulation. This combined tool is able to compute robust assembly paths for industrial robots, i.e. paths less sensitive to the geometrical variation existing in the robot links, in its control system, and in the environment. This may lead to increased productivity and may limit error sources. The main idea to improve robustness is to enable robots to avoid motions in areas with high variation, preferring instead low variation zones. The method is able to deal with the different geometrical variation due to the different robot kinematic configurations. Computing variation might be a computa- tionally expensive task or variation data might be unavailable in the entire state space, therefore three different ways to estimate variation are also proposed and compared. An industrial test case from the automotive industry is successfully studied and the results are presented. © 2013 The Society of Manufacturing Engineers. Published by Elsevier Ltd. All rights reserved. 1. Introduction A common scenario when simulating manufacturing and assembly processes includes an engineer trying to simulate the process by manipulating objects in a digital mock-up software. In highly geometrical restricted assembly situations, this procedure is often sensible to errors and is time consuming. Furthermore, it is common that such manufacturing and assembly tasks are performed by robots, whose motions are difficult to plan and control by manual programming. If we also consider that, in real- ity, every physical object is subject to geometrical variation due to its manufacturing process, then it appears prohibitive for an engineer to verify the feasibility of the assembly procedure at an early stage. An automated verification is therefore helpful, since it can decrease the enormous costs that arise when realizing the infeasibility of an assembly plan late in the production phase, and the following need to re-design the process and/or the prod- ucts. Robots performing assembly operations are subject to variation as any other assembly system. Another variation source comes from ∗ Corresponding author. Tel.: +46 317724252. E-mail address: domenico.spensieri@fcc.chalmers.se (D. Spensieri). robot resolution, mainly due to the precision of the computing sys- tem and to sensors and actuators sensitivity. Resolution affects also the accuracy of the robot, which is a measure of how close to the nominal value the robot can reach. The accuracy, then, influences the repeatability, which is the ability of a robot to perform the same task in the same manner, see [1]. 1.1. Motivation One way to substantially improve positional accuracy is by on- line teaching the robot the poses it will assume during its assigned operations: in this way the robot controller stores its internal state and the data on how to perform the same task in the future, see [2] for advances in on-line programming environments. Anyway, pro- gramming robots on-line in order to perform tens or hundreds of tasks, with their respective paths and via-points, can be prohibitive. Another way to achieve more accurate programs is by robot cal- ibration: during this operation the mechanical parameters of the robot model are identified, see [1,3], e.g. by measuring the differ- ences between the estimated position of the Tool Center Point (TCP) and the actual one. Anyway, the results may vary depending on which area of the workspace the calibration is done, on the robot’s load, as well as on speed and acceleration of the motion. Moreover, it is difficult to compensate for uncertainties in the robot’s degrees 0278-6125/$ – see front matter © 2013 The Society of Manufacturing Engineers. Published by Elsevier Ltd. All rights reserved. http://dx.doi.org/10.1016/j.jmsy.2013.04.013