See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/261021642

Automatic Path and Trajectory Planning for Robotic Spray Painting

Conference Paper · January 2012

CITATIONS READS

41 1,896

4 authors, including:

Alessandro Gasparetto Renato Vidoni

University of Udine University of Udine
200 PUBLICATIONS 3,848 CITATIONS 166 PUBLICATIONS 3,448 CITATIONS

SEE PROFILE SEE PROFILE

Daniele Pillan
CMA Robotics
6 PUBLICATIONS 120 CITATIONS

SEE PROFILE

All content following this page was uploaded by Daniele Pillan on 13 September 2021.

The user has requested enhancement of the downloaded file.

 ROBOTIK 2012

Automatic Path and Trajectory Planning for

Robotic Spray Painting.
Topic: Robotics in Production

Alessandro Gasparetto,
Dep. of Electrical, Mechanical, and Management Eng. University of Udine, Universitá di Udine, Italy

Renato Vidoni
Faculty of Science and Technology, Free University of Bozen-Bolzano, Italy

Daniele Pillan
CMA-Robotics, Pradamano (Udine), Italy

Ennio Saccavini
CMA-Robotics, Pradamano (Udine), Italy

Abstract

In this paper, a new automated optimum path and trajectory generation system for robotic painting process is presented.
Usually, a direct self-learning manual method is usually adopted, i.e. the operator drives the robot manually through a
complete spraying cycle. Recently, in order to-avoid this time-consuming procedure, CAD-based or "acquire-compare-
recognize" methods have been developed. Here, a general technique that avoids the need for manual programming or
CAD drawings and allows to obtain an optimal path and trajectory by using the graph theory and operative research
methods is developed. After an image acquisition phase, a partitioner splits the object into a set of primitives, then the
proposed algorithm is run on the graph in order to generate the optimal path; then, a optimum trajectory planning phase is
performed. Kinematic and dynamic simulators of the CMA Robotics robots have been developed and validated in order to
allow a correct planning and evaluation of the results. The new path planning algorithm has been implemented in Matlab
and Visual-Studio.NET environments. Practical applications are presented.

1 Introduction ods that realize optimal spray gun paths or trajectories,

thus avoiding the manual self-learning programming pro-
The spray painting production is widely used in industry cedures, have been developed. Computer Aided Design
and becomes everyday more competitive. Current research (CAD) tool path planning methods are extensively used in
deals mainly with automation of the process, optimization automotive industry. A model and pieces CAD-drawing
and quality of the paint. Indeed, the uniformity of the paint [1, 2, 3] are directly exploited to create a path for the gun
thickness on a product strongly influences its quality and and product manufacturing process. This solution, that
such a feature is typically determined by the robot tool allows to reduce dramatically the operation time and the
path. As a consequence, in spray painting processes, tool human labor, needs a CAD-drawing of the pieces that is
path and trajectory planning with uniform material distri- not always available in many industrial applications (e.g.
bution and minimum wasted paint remain challenging re- painting of windows, doors, chairs, home appliances).
search topics. Thus, it is rarely adopted in small industries also because
such CAD software packages can be expensive and no
In order to specify a path for a robotic manipulator, a man-
trained and qualified personnel is available.
ual tool planning, i.e. a teaching method, is usually per-
formed. An operator teaches the robot painting path by A different approach for the automation of the painting
moving the end-effector around the parts to be painted and, process deals with the exploitation of a vision system: the
in the meantime, positions and orientations are recorded by image of the piece is acquired and matched with one of the
the robot controller. The overall procedure requires a lot of patterns stored in a dynamically updated library [4, 5, 6].
time. In order to obtain a good painting result, the operator The technical challenge is the detection of the geometry of
has to carry out extensive tests on a work cell and the final the part and the automatic generation of the robotic paint-
path is often the result of a trial and error approach. ing path and trajectory. When small different lots or par-
In the last years, automatic and semi-automatic meth- ticular and new shapes are to be painted (e.g. customized

211
doors, windows and gates), such a method can fail since Figure 1: Layout of the (semi)-automated process
no pre-loaded forms are present in the library and errors
occur.
In this work, the improvement of the efficiency of the
CMA-robotics painting process has been evaluated in order 1.1 Image acquisition
to generate continuous and smooth motions of the spray
paint gun by taking into account the optimization of the To define the structure of the object to be painted automat-
overall painting path and trajectory definition. ically, i.e. detect the edges and extract primitives, pattern
recognition techniques and/or artificial vision can be used.
The robotized painting process and system developed in
In this work, objects are detected using a barrier sensor
this work exploits robot vision to define the structure of
which working principle is based on the interruption of the
the object to be painted, in motion on a conveyor, by de-
light beam (see Fig.2). These sensors are independent from
tecting its front edge and primitives in an automatic way.
the reflection and from the color of the object and, thus, are
Thus, thanks to a partitioning phase, the issue of how to
the ideal solution for this kind of task also taking into ac-
find an optimal painting path is addressed studying a col-
count the fact that objects are in motion on a conveyor.
lection of basic geometries. In such a way, any kind of
shape can be treated and an optimal aesthetic result and a
minimization of the wasted paint can be reached by plan-
ning a continuous and smooth robot motion. This search
for the optimal path is done by means of the set of basic
geometric coordinates result of the partitioning phase and
of the representation of the object as a graph.
A graphical interface allows the user to choose the opti-
mal path between different available solutions in an auto-
matic or semi-automatic way. These solutions are related
to different criteria such as painting priority (e.g. internal Figure 2: Barrier sensor.
or external elements first, plans or lines), path continuity
and optimal fluidity. The obtained path is tested using a At the end of the acquisition phase a binary image is avail-
simulator, integrated into the developed system, that im- able. This is then elaborated by a software module able to
plements a model of the anthropomorphic robot to which recognize objects and their contours and surfaces. Indeed,
the painting task is assigned. this partitioner extracts from each distinct object a set of
primitives (Fig. 3), defined as the different simple entities
The availability of the path allows to plan the trajectory by
that constitute the object.
considering or setting velocity, spray gun opening/closing
Two main kinds of primitives are recognized: segments or
phases, spray distance, flux and color of the paint. Since
curves that form the edge and segments, curves and planes
constant speed is advisable to obtain a better quality paint-
called “frontal” primitives.
ing, trapezoidal speed profile or polynomial primitives that
allow at least the continuity of the velocity are chosen. The width of the painting stroke defines the limit between
segments and planes. Indeed, a plane consists of a rectan-
Thanks to a dynamic simulator of the robot, the overall
gle with both sides longer than the painting stroke width
planned trajectory can be evaluated and the required en-
and is composed of segments of maximum width parallel
ergy consumption can be forecast and eventually mini-
to the plane longer edge. The direction of the strokes is car-
mized. In Fig.1 the structure of the automated process is
ried along the longer side of the plan or alternatively can
presented and in the following the main phases of the pro-
be chosen by the user (vertical or horizontal). The distance
cess are described.
between the strokes is determined by interface graphics.
A file containing the characteristics of the object is then
stored and sent to the Mapping phase.

Figure 3: Image elaboration

212
1.2 Mapping and searches the primitive belonging to the edge. After
that, the graph is subdivided in subgraphs constituted from
Mapping is constituted by the primitives of the object. The primitive that guarantee the continuity of the path and the
object to be paint is considered as a graph [7], a useful rep- imposed requisites. Then, optimization is searched. Dif-
resentation because it allows to easily determine the conti- ferent requests con be chosen:
nuity of the primitives. Indeed, segments or curves are con-
sidered as arcs and the coordinates that constitute them are
nodes. Graphs can be represented extensively as a list of all • Contour and minimum angle priority (see Fig.4).
the elements, i.e. nodes and vertexes. Different represen- This function imposes a priority in the execution of
tations are possible, e.g. adjacency matrix, adjacency list, the contour primitives and, if more than one contour
incidence matrix, path matrix, circuit matrix, cut matrix arc insist on a node, the one that forms an angle near-
and distance matrix. A constraint of the project consists est to l80◦ with the previous primitive is selected.
of avoiding multiple passes along the same primitive, it is This function is cyclically repeated until all the edge
therefore important to use a representation that underlines primitives are considered.
the arcs. A B-type matrix of incidence representation has
been chosen in order to easily evaluate if multiple passes • Contour and maximum length priority. This function
along the same primitive are planned. The incidence ma- imposes a priority in the execution of the contour
trix has dimensions n × m where the number of rows n is primitives privileging the maximum path length.
equal to the number of nodes and the number of columns
m is equal to the quantity of arcs. Each arc is represented
in the columns of the incidence matrix with two elements • Path continuity and minimum angle priority. This
that assume a value equal to one in correspondence to the function privileges the path continuity starting from
nodes that constitute. Clearly, the sum of the columns of the contour primitives.
an arc is equal to two.
• Path continuity and maximum length priority. This
function recursively searches the primitives that al-
1.3 Path and Trajectory Planning low the maximum length path.
An optimal path for the painting of an object of any form
and dimension that satisfies the requisites imposed by the • Maximum length sub-graph. This function searches
practical applications is the result of this phase. the primitive sequence that forms the maximum
The project requisites depend on the application. The most length path. If the same maximum length is reached
important and common are: cover the whole figure, avoid by more than one sub-graphs, the one with the first
manifold passages along the same primitive, allow con- node nearest to the chosen starting point of evalua-
stant speed during the painting, allow the choice of the tion is selected.
starting point and search continuity of painting provided
that the angle formed by two primitives with an extreme in
common is greater than a critical fixed angle. Specific settings can be chosen by the operator and auto-
Among the algorithms proposed by the operational re- matically checked when optimal path is searched. Among
search (e.g. shortest path, minimum length algorithms these critical angle, end-effector distance from the edge of
[8, 9]) it is hard to find out a solution suitable for all the the object at the beginning and end of the path, z axis and
previously mentioned constraints. Euler angles of the spray gun, search mode for internal and
Eulerian cycles (Chinese Postman Problem) could be the edge primitives are the most important.
optimal solution, indeed an Eulerian path is a path in a Finally, the optimal path found is stored in a file and sent to
graph which visits each edge exactly once but for having the trajectory planner module. The trajectory is planned at
an Eulerian path the graph must be connected and has to constant speed in order to obtain a better quality painting.
show particular features. Thus, in order to have a plan- For this reason a trapezoidal speed profile which requires
ner independent from the form of the considered object, constant acceleration in the starting phase, a cruising con-
from the number of incidental arcs in every node and from stant speed and a constant deceleration in the final phase is
the choice of the type of optimal path (maximum path or usually adopted.
search of the fluidity), an ad-hoc iterative algorithm based The spray is modeled as a cone with an ellipse base.
on the graph theory has been realized using the character- The parameters of the ellipse are calculated from the
istics and the requisites of the project (see Fig. 4 as an fan parameters given by the gun constructor. The value
example). A search process that specifies which rules ap- of the cone height corresponds to the spray distance
ply and when to stop the calculation if a termination con- from the gun to the object to be painted. To opti-
dition is met has been implemented in Matlab and Visual mize the time of painting off all the objects the al-
Studio.Net environments. Firstly the algorithm find the gorithm finds the minimum length in transition from
nodes of the segments and arcs that constitute the object one component to another and from different objects.

213
 In order to show the effectiveness of the dynamic
simulator, a comparison between simulated and mea-
sured torques for the second motor of the anthro-
pomorphic robot under a painting trajectory is pre-
sented in Fig. 5. As can be seen, the simula-
tor keeps well the dynamics of the real system and
thus, the simulated results can be considered as realistic.

Figure 5: Simulated and experimental torque comparison:

red (simulated); experimental (green)

Moreover, the simulator module uses the same interface

present on the programming devices of the robots. A mesh
creator converts the digital image, that contains the objects
to be painted, in a polygon meshes. Meshes are a collec-
tion of vertexes and faces that define the shape of an object
and simplify the graphic rendering.
In the simulation, the quantity of paint applied on the
whole surface of the object is highlighted by means of the
variation of different colors. In Fig. 6 the graphical inter-
face of the simulation module is presented.

Figure 4: Contour and minimum angle priority algorithm.

1.4 Kinematic and dynamic models and sim-

ulators
The optimal path and trajectory are sent to the ad-hoc de-
veloped simulator module that reproduces the kinematics
and dynamics of the anthropomorphic robot to which the
painting task is assigned.
Key concepts of robotics [11] such as direct kinematics, in-
verse kinematics, differential (first and second order) kine-
matics, statics and dynamics have been evaluated and im-
plemented. The dynamic simulator has been developed Figure 6: Graphical interface of the simulator module
taking into account the geometrical and inertial parameters
of the robots in use; in particular, a Newton-Euler approach The path of the end-effector can be simulated highlighting
has been implemented. Thanks to this simulator, a direct the strokes. Alternatively, the painting under examination
evaluation of the required effort for the chosen trajectory can be reproduced by highlighting the on/off spray gun and
can be performed. the object lines painted.

214
2 Implementation
The overall idea has been extensively tested and validated.
First of all, the algorithm has been experimented on a CMA
Robotics’ robot, the GR-6100: an anthropomorphous ma-
nipulator suitable for painting in line with the spray gun
on the end-effector. The basic steps of the process have
been performed: artificial vision acquisition, partitioning,
mapping, path searching, trajectory planning and dynam-
ics and painting simulation, and, finally, physical painting.
The object is moved by means of an aerial conveyor at 5
m/s; the optical barrier resolution is less than 1 cm. The
developed algorithm requires less than one second to ob-
tain a solution (about 0.62 seconds with a PC Intel Core 2 Figure 8: Painting phase
Duo 2.40GHz, ram 4GB).
Usually, the manual self learning method takes minutes
for moving the robots end-effector around the piece to be
painted, to store points and create the program, while the 3 Conclusions
developed system takes few seconds. Moreover, the path
planning step allows the user to select different optimal An automated optimum path and trajectory generation sys-
paths without a process interruption. Thus, the overall pro- tem for robotic painting process is presented.
cess is much more efficient with respect to the traditional The process structure, the automation and the implemen-
manual one. tation steps are described. Graph theory and operative re-
After the experimental validation, the system has been im- search techniques are applied to provide a general and op-
plemented and integrated into the CMA-Robotics painting timal solution for the path planning problem. The path op-
systems. Fig.s 7 and 8 are taken from a developed applica- timization has been done considering the coordinates of
tion. the primitives such as nodes, segments and curves like arcs
and creating an algorithm able to take into account several
constraints posed by robotized painting.
The system is easy-to-use and flexible allowing to treat dif-
ferent and special shapes, and to fill the gap between the
image acquisition and trajectory planning process leading
a new, fast, general and automatic mechatronic system for
robotic painting purposes.

References
[1] Li, X., Landsnes, Ø.A., Chen, H., Sudarshan, M-V,
Fuhlbrigge, T.H., Rege, M-A., Automatic Trajectory
Generation for Robotic Painting Application In Proc.
of Robotik 2010.

[2] Chen, H., Fuhlbrigge, T., and Li, X., Automated in-
dustrial robot path planning for spray painting pro-
cess: A review In Proc. of the 4th IEEE Conference
on Automation Science and Engineering, 2008.

[3] Chen H, Fuhlbrigge T, Li X, A review of CAD-based

robot path planning for spray painting. Industrial
Robot: An International Journal, Vol. 36 Iss: 1, pp.45
- 50, 2009.

[4] Vincze, M., Pichler, A., Biegelbauer, G., Husler, K.,

Andersen, H., Madsen, O., and Kristiansen, M., Au-
tomatic robotic spray painting of low volume high
variant parts. In Proc. of the 33rd ISR: International
Figure 7: Image acquisition phase Symposium on Robotics, 2002.

215
 [5] Vincze, M., Biegelbauer, G., and Pichler, A., Paint- [8] Sebastian Thrun, Wolfram Burgard, Dieter Fox Prob-
ing parts automatically at lot size one. In Proc. of abilistic Robotics, Mitt Press, Massachusetts 2005,
the International Workshop on Robot Sensing, 2004
(ROSE 2004), 2004. [9] Mauro Cerasoli, Franco Eugeni, Marco Protasi Ele-
ments of discrete mathematics, Zanichelli, Bologna
[6] Pichler, A., Vincze, M., Andersen, H., Madsen, O., 1992,
and Hausler, K., A method for automatic spray paint-
ing of unknown parts. In Proc. of the 2002 IEEE In- [10] Edward Minieka Optimization Algorithms for Net-
ternational Conference on Robotics and Automation works and Graphs, Wilbur Meyer, New York 1978,
(ICRA ’ 02), 2002.
[11] Bruno Siciliano, Lorenzo Sciavicco, Luigi Villani,
[7] University of Liverpool, Graph theory in operations Giuseppe Oriolo, Robotics: Modeling, Planning and
research, T.B. Boffey, Hong Kong 1982, Control, Milan, McGraw-Hill 2008

View publication stats

216