Mech. Sci.

, 14, 87–97, 2023

https://doi.org/10.5194/ms-14-87-2023
© Author(s) 2023. This work is distributed under
the Creative Commons Attribution 4.0 License.

Obstacle-avoidance path planning based on the

improved artificial potential field for a 5 degrees of
freedom bending robot
Quansheng Jiang1 , Kai Cai1 , and Fengyu Xu2
1 School of Mechanical Engineering, Suzhou University of Science and Technology, Suzhou 215009, China
2 College of Automation, Nanjing University of Posts and Telecommunications, Nanjing 210023, China
Correspondence: Quansheng Jiang ([email protected])

Received: 2 October 2022 – Revised: 25 December 2022 – Accepted: 30 January 2023 – Published: 24 February 2023

Abstract. Path planning is a key technique used in the operation of bending robots. In this paper, an obstacle-
avoidance path-planning method of a 5 degrees of freedom (5 DOF) bending robot based on improved artificial
potential field is proposed. Firstly, a connecting-rod coordinate system of the 5 DOF Cartesian bending robot
is established to determine an equation of motion trajectory of the bending robot. Secondly, in view of the
problem of the local minimum in the artificial potential field (APF) method and the failure of path planning, an
improved APF path-planning method based on a rapidly exploring random tree (RRT) algorithm is proposed,
which reduces the length of the path and enhances path smoothness. Finally, through simulation and obstacle-
avoidance experiments on the path of a mechanical arm, effective path planning based on the improved APF
method is verified. The experimental results show that the proposed path-planning method can plan an optimal
path and meet the technical requirements of bending robot operations.

1 Introduction from the starting point to the target point, but it is easy to fall
into the local optimal solution.
In the industrial production field, bending robots replace hu- Environment modeling and path search are two important
man labor in bending operations, which greatly improves issues in global path planning. Global path-planning methods
working efficiency and machining quality. Path planning in include the visibility graph, A∗ algorithm, D ∗ algorithm, and
the machining process is a key technical link in the per- artificial potential field (APF) (Shao et al., 2019; Wang et al.,
formance delivery of bending robots. Its task is to find a 2019; Rasekhipour et al., 2016). Other global path-planning
continuous path from the starting point to the target point methods are intelligent optimization algorithms, such as ge-
in the set area and avoid obstacles under the optimal or netic algorithm and particle swarm optimization (Korayem et
quasi-optimal performance index (Akbaripour and Mase- al., 2016; Kim and Lee, 2015). Using a differential evolution
hian, 2017; Stentz, 1994). algorithm, Das et al. (2018) propose a method of energy-
According to the degree of environmental information, the efficient path planning that can be applied to an industrial
path-planning methods of mobile robots are usually divided robot arm in a workspace containing multiple obstacles. A
into global path planning and local path planning. Global suitable cost function is used to track the optimal path be-
path planning is based on the known global obstacle map in- tween initial and final configurations of the robot arm joints.
formation and uses various optimization algorithms to obtain For industrial robots, exploring in an unknown envi-
the global optimal path, but it has some disadvantages such ronment is still a challenging task. In terms of percep-
as large computation, low search efficiency, and slow conver- tion and positioning of obstacles, Gala et al. (2018) and
gence speed. Local path planning can acquire the information Gala and Sun (2019) propose a three-dimensional sound
of the current position and the local obstacle in real time ac- source localization technology based on a self-rotational
cording to its own sensors, so as to obtain the optimal path two-microphone array. It can estimate the position of a sta-

Published by Copernicus Publications.

88 Q. Jiang et al.: Path planning for a 5 DOF robot

tionary sound source in a spherical coordinate system. Fil- duced the length and turning angle of the path. Li et al. (2017)
ipenko and Afanasyev (2018) compare various simultane- proposed a trajectory planning method for sprawling robots
ous localization and mapping (SLAM) systems of mobile inspired by a trotting animal and offered references to the
robots in an indoor environment. The advantages and disad- trajectory planning of robots.
vantages of each SLAM system in different application sce- In the path-planning algorithms for obstacle avoidance, it
narios are analyzed. Zheng (2019) introduces a new method is common to use the APF method to abstract details per-
to investigate the trajectory planning for the multiple degrees taining to the actual space. In reference to this, Cheng et
of freedom (MDOF) robot in improving accuracy and stabil- al. (2019), based on the task of path planning of a mo-
ity. Fan et al. (2020) present a concise path-planning method bile robot, improved the traditional A∗ algorithm and APF
for an autonomous underwater vehicle (AUV) based on an method to shorten the driving path of the mobile robot. It
improved artificial potential field (APF) method, which con- solved the problem of path-planning failures in the case of a
siders the spatial location and direction of the velocity of the local minimum being located. To improve the global search
moving objects. Cong et al. (2020) present a bidirectional ability of path planning for a mobile robot, the improved APF
fast extended random tree algorithm rapidly exploring ran- and ant colony algorithm (ACA) under a global static envi-
dom tree connect (RRT) to complete the dual-arm path plan- ronment was proposed in reference to Wang et al. (2018),
ning under the planning framework of “MoveIt”. which allowed planning of the optimal path, albeit with com-
Considering a safe distance between an end effector of plex computational processing.
a robot and obstacles to avoid collision, Lai et al. (2017) Path smoothing is an important part of path planning;
presented an obstacle-avoidance method for a mechanical the path obtained by the traditional planning algorithms is
arm based on non-uniform rational B-spline curves. Li et usually a sequence of multiple line segments and is a non-
al. (2022) proposed an improved A∗ algorithm by a bidi- smooth curve, which readily causes discontinuous motion of
rectional alternating search strategy for mobile robots path a robot, while a smooth path can ensure continuous motion.
planning. Nishi and Mori (2018) proposed a motion planning In this paper, an improved APF-based path-planning method
method based on the potential field, which saves energy and based on the idea of RRT is proposed, which reduces the
optimizes picking points of the workpieces. Contreras-Cruz length of the path and makes the path smoother.
et al. (2015) proposed a robust algorithm based on meta- The remainder of this paper is organized as follows. In
heuristic methods to generate a path composed of forward- Sect. 2, the kinematics of the 5 DOF bending robot is ana-
ing motion and rotation. By adding a singular value decom- lyzed. In Sect. 3, trajectory planning of the robot based on
position method to optimize the measurement of operability, the improved APF method is introduced. Simulation and ex-
Ahmed et al. (2022) proposed an integrated path-planning perimental results are presented and discussed in Sect. 4, re-
algorithm based on the deformation of thin metal plates. spectively. Section 5 concludes the paper.
Kovács et al. (2016) presented a mobile robot path-planning
method using animal motion attributes, and it was capable of
2 Kinematic analysis of the 5 DOF bending robot
online navigation in dynamic environments.
In terms of path planning based on machine learn- 2.1 Establishment of the coordinate system of the
ing, Zhou et al. (2022) developed an improved lazy 5 DOF robot
probabilistic-roadmap-algorithm-based online collision-free
path-planning method for arc welding robots. Zhuang et For a bending robot, it is necessary to find feasible solutions
al. (2019) proposed an RRT algorithm with a variable step in high-dimensional configuration space. Therefore, a spa-
length for path planning of a dual-arm robot, which improved tial coordinate system needs to be established for the bend-
search efficiency and shortened the time required for path ing robot to obtain the spatial position of its end effector,
planning. Yu et al. (2017) proposed a task-oriented simula- thus turning path planning into a constrained optimization
tion system of a mechanical arm based on a recurrent neural problem (Ratiu and Prichici, 2017). A mechanical arm of the
network and obtained a mapping trajectory between angles of bending robot is a multi-DOF structure with multiple rigid
joints of the robot arm. By a dynamic fuzzy neural network bodies connected in series. To describe changes in the rela-
algorithm, Han et al. (2015) obtained inverse kinematic pa- tive position between adjacent connecting rods, a coordinate
rameters of a robot, which improved the calculation accuracy system is established for each connecting rod, thus obtaining
and efficiency. Jun et al. (2017) proposed an improved spline position coordinates of the end effector of the mechanical
interpolation method based on spline functions and genetic arm in a three-dimensional (3-d) coordinate system.
algorithms, which solved the problems of complex workload During the motion of the mechanical arm, the coordinate
and inflexible operation in path planning. Wang and Wan system of the base remains fixed, and the relative positions
(2015) used the improved quintic polynomial interpolation of other connecting-rod coordinate systems are ascertained
method in a joint space. Sun et al. (2017) presented an im- based thereon. The 5 DOF Cartesian bending robot consists
proved A∗ algorithm for global path planning and divided of three moving joints and two rotating joints connected in
the planned path according to a small step length, which re- series. For three joints moving in a straight line, the Z axis

Mech. Sci., 14, 87–97, 2023 https://doi.org/10.5194/ms-14-87-2023

Q. Jiang et al.: Path planning for a 5 DOF robot 89

Figure 3. The motion coordinate system of the bending robot fol-

lowing the bending machine.

part. After the host computer sends the bending command,

the sliding blocks of the bending machine start to operate
from the top dead center. When the tool tip of the upper dies
contacts the materials to be bent, the bending machine sends
Figure 1. Coordinate system of 5 DOF rectangular coordinate ma-
the bending command to the robot. According to the planned
nipulator. path algorithm, the robot controls the mechanical arm to fol-
low the bending process. In the motion-following process
for bending the mechanical arm, the Y axis, Z axis, and A-
rotation axis of the bending robot need to follow the bending
of the sheet-metal part. The coordinate system of the robot
following the bending machine is shown in Fig. 3.
In Fig. 3, the coordinate system XOY is established with
the end effector of the bending robot at the center, while the
coordinate system X0 O0 Y0 is a basic coordinate system of the
bending machine. I , d, and α indicate the thickness of the
end effector of the robot, the thickness of the sheet-metal
part, and the angle between the sheet-metal part clamped by
the end effector of the mechanical arm and the X axis during
the bending operation, respectively. L, β, W , and S repre-
sent the horizontal distance from the reference origin O of
the sheet-metal part to the center O0 of a V -shaped groove,
the angle of the sheet-metal part to be bent, the width of the
Figure 2. Coordinate system of 5 DOF Cartesian robot. V -shaped groove in the lower die, and the bending depth of
the sheet-metal part, respectively. I , d, and L are constants,
while α varies. When calculating the bending about the Y
denotes the direction of movement of each joint. For the two and Z axes and the A-rotation axis of the bending robot fol-
rotating joints, the Z axis represents the rotational direction lowing the bending of sheet-metal part, it is only necessary
of each joint. The first three moving axes determine the po- to calculate the kinematic equations pertaining to the Y and
sition of the end effector, and the last two rotation axes de- Z axes of the robot relative to the angle α.
termine the pose of the end effector. The X axis denotes the The angle between the sheet-metal part clamped by the
Denavit–Hartenberg (D–H) modeling rule, and the Y axis is end effector of the mechanical arm and the X axis shows the
determined according to the right-hand rule; therefore, the following relationship with bending depth S and width W of
established connecting-rod coordinate system of the 5 DOF the V -shaped groove (known constants):
Cartesian mechanical arm is shown in Fig. 1. The established !
coordinate system of the 5 DOF Cartesian bending robot is S
shown in Fig. 2. α = arctan W

. (1)
2

2.2 Kinematic analysis of the bending robot The relationship between α and the angle β of the sheet-
metal part to be bent is expressed as follows:
This bending robot, with 5 DOF, cooperates with a computer
numerical control (CNC) bending machine to bend a sheet- 180 − β
metal part. An end device of the robot clamps the sheet-metal α= . (2)
2

https://doi.org/10.5194/ms-14-87-2023 Mech. Sci., 14, 87–97, 2023

90 Q. Jiang et al.: Path planning for a 5 DOF robot

The bending depth S is given by where U (i) indicates the virtual potential field; and Uatt (i)
W 180 − β W and Urep (i) represent virtual potential fields of the attractive
S= tan = . (3) and repulsive, forces, respectively.
2 2 2 tan β2 If the bending robot moves in a two-dimensional finite
Similarly, according to the geometric relationship of moving space and its position in the motion space is Pi , the potential
coordinates in Fig. 3, the displacement of point O in the co- function of attractive force between the robot and the objec-
ordinate system of the robot’s end effector along the Z axis tive point is expressed as follows:
is obtained as follows: 1 2

!
W Uatt = ka Pi − Pg , (8)
2 2
Zo0 = I + d + L − sin α − I cos α. (4)
cos α
where ka and Pi − Pg denote the scale factor of attractive
The displacement of point ‘O in the coordinate system of the force and the distance between the current position and the
robot’s end effector along the Y axis is objective position of the robot, respectively.

! The potential function of the repulsive force between the
W
2 W bending robot and obstacles is written as follows:
Yo0 = L − L − cos α − I sin α −
cos α 2 ( 2
1 1 1
= L(1 − cos α) − I sin α. (5) Urep = 2 k r di − d0 di ≤ d0 , (9)
0 di > d0
Bending equations about the Y and Z axes of the robot with
respect to angle α are where kr , di , and d0 indicate the scale factor of the repulsive
 ! force, the distance between the robot and obstacles, and the
S radius of influence of the potential field of the repulsive force

α = arctan  W 



of obstacles.

2


Yo0 =L(1 − cos α) − I sinα! . (6) When the negative gradient energy in the potential fields of
W attractive and repulsive forces acts on the robot, the resultant


2

 Zo0 = I + d + L − cos α sin α − I cos α

force on the robot is expressed as follows:



Based on the above bending process, the motion path of F (i) = −∇U (i) = −∇Uatt (i) − ∇Urep (i), (10)
the end effector of the bending robot can be determined
through calculation. In the joint space and the Cartesian where ∇U (i) represents the gradient of the potential field at
space, the bending trajectory of the robot can be separately the point i.
planned. The attractive and repulsive forces applied to the robot can
be separately expressed as follows:
3 Trajectory planning of the robot based on the


Fatt (i) = −ka Pi − Pg , (11)
improved APF method

3.1 Artificial potential field (APF) method (  

1
kr − d10 1 ∂d
di ≤ d0
The APF method is a virtual force method that constructs the Frep (i) = di di2 ∂X . (12)
fields of attractive force and repulsive force acting together 0 di > d0
around the objective position and obstacles by introducing
The APF method searches for the optimal solution by simu-
a potential field function to describe the workspace of the
lating the energy field, but there are two special cases: one is
robot (Orozco-Rosas et al., 2019). The robot is subject to the
that when the objective point is within the influencing range
joint action of the attractive force generated at the objective
of the repulsive force of obstacles, the repulsive force will be
point and the repulsive force generated by obstacles in the
very large and the attractive force is relatively small, which
virtual potential field, and the optimal collision-free path is
makes the objective point unreachable. The other one is that
planned by searching the descent direction of the potential
when the robot is in a straight line with the objective point
function. The APF method, with its high computational ef-
and obstacles, or it encounters U -shaped obstacles during
ficiency, can achieve real-time path planning and trajectory
motion, it will be trapped in a local minimum.
smoothing; however, it is often trapped in local optima, pre-
In view of the above shortcomings of the APF method, a
venting path planning (Wan et al., 2019; Kovács et al., 2016).
new potential field function is constructed for improvement,
The energy applied to the robot at a certain point in the
based on the idea of the RRT algorithm when the local po-
trajectory can be expressed as
tential model conflicts with the global one during the path
U (i) = Uatt (i) + Urep (i), (7) selection of the robot.

Mech. Sci., 14, 87–97, 2023 https://doi.org/10.5194/ms-14-87-2023

Q. Jiang et al.: Path planning for a 5 DOF robot 91

3.2 Improved APF method based on the RRT algorithm the extension is shown in Fig. 5. The idea is to add an
objective function G(n) of attractive force to each node
The RRT algorithm is a search algorithm based on a tree n in the random tree (node n represents the nth node
structure. It is a planning algorithm that extends outward xnew as the random tree extends outward from the ini-
into a tree structure from the initial point and randomly sam- tial point xinit ); the objective function of attractive force
ples points in the planning space to determine the next direc- can be expressed as follows:
tion of extension, thus realizing rapid path planning. Char-
acterized by the advantages of avoiding the physical mod- xgoal − xnear
G(n) = ρka , (13)
eling and mathematical modeling of the objective space, xgoal − xnear
quickly expanding to explore the feasible region in the
where ρ and ka indicate the step length and scale fac-
space, and providing efficient execution, the method has been
tor of attractive force, respectively; and xgoal − xnear
widely used.
represents the absolute value of the geometric distance
The main idea of the improved APF-based path-planning
between node n and the objective point xnear .
method based on the RRT algorithm is to improve the APF
method with the random extension and search characteris- The RRT algorithm of a new node xnew can be expressed
tics in order to avoid local minima. When falling into the by the following equation:
local minimum, the RRT algorithm is used to generate uni- xrand − xnear
form numbers of root nodes around obstacles, and then the R(n) = ρ . (14)
kxrand − xnear k
components of attractive and repulsive forces are applied to
the root nodes to generate a local random tree near the local The equation for generating a new node after introduc-
minimum. Under the guidance of components of attractive ing the component of attractive force into the RRT al-
and repulsive forces in the APF, the random tree grows to the gorithm is shown as follows:
objective point, thus solving the problem, such as that asso- xgoal − xnear xrand − xnear
ciated with an unreachable objective point and local minima xnew = xnear + ρka +ρ . (15)
xgoal − xnear kxrand − xnear k
found during motion planning.
In this way, the points in the random tree can search for
1. Search process using the RRT algorithm. and grow to the objective direction in the free space un-
der the action of the component of attractive force, thus
The specific search process using the RRT algorithm
avoiding the problem that the objective point is unreach-
is as follows: an initial point is selected in the envi-
able by the robot.
ronment, which is recorded as a root node xinit of the
random tree. By searching for the free space, a random 3. Extension of the RRT algorithm by increasing the com-
sampling point xrand is selected. If xrand is not in the ponent of repulsive force.
region with obstacles, xinit and xrand are connected to The component of repulsive force is added to the RRT
obtain a line L. If the whole line L is not in the region algorithm to guide the local random tree to grow away
with obstacles, a new node xnew is generated along line from the obstacles. The extension diagram is demon-
L from xinit to xrand according to a certain step length ρ. strated in Fig. 6. The core idea is to add a function T (n)
In this way, the simplest tree is formed by xinit , xnew , and of repulsive force of obstacles into each node n in the
their connection. By selecting node xnear nearest the ran- random tree, where node n represents the nth node xnew
dom sampling point xrand , they are connected, and a new as the random tree extends outwards from the initial
node xnew is generated from xnear at a certain step length point xinit . The function of the repulsive force of ob-
ρ. If there is no collision with obstacles during exten- stacles can be expressed as follows:
sion from xnear to xnew , a random tree is generated by (  
adding this new node xnew into the random tree. When ρkr 1
− 1 1 ∂(xnear −xobstacle )
pn ≤ p0
T (n) = pn p0 pn2 ∂xnear , (16)
sub-nodes in the random tree contain the objective point 0 pn > p0
xgoal , a path from the initial point xinit to the objective
point xgoal can be generated in the random tree. If col- where ρ and kr denote the step length and scale fac-
lision occurs, this extension is canceled. A schematic tor applied to the repulsive force, respectively; pn de-
representation of the path search in a space containing notes the shortest distance from the node to obstacles;
obstacles using the RRT algorithm is displayed in Fig. 4. and p0 and xobstacle denote the influence range of re-
pulsive force of obstacles on the node and the position
2. Extension of the RRT algorithm by increasing the com- vector of obstacles.
ponent of attractive force. The RRT algorithm of the new node xnew can be ex-
The component of attractive force is added to the RRT pressed by the following equation:
algorithm to guide the local random tree to grow to- xrand − xnear
wards the objective point, and a schematic diagram of R(n) = ρ . (17)
kxrand − xnear k

https://doi.org/10.5194/ms-14-87-2023 Mech. Sci., 14, 87–97, 2023

92 Q. Jiang et al.: Path planning for a 5 DOF robot

Figure 6. Extension map of RRT algorithm with added repulsion

component.

Figure 4. Schematic diagram of RRT algorithm searching in obsta-

cle space.

Figure 5. Extension map of RRT algorithm with added gravity

component.

The equation for generating a new node after introduc-

ing the component of repulsive force into the RRT algo-
rithm is given by
  

 xnear + ρkr p1n − p10 p12

 n

xnew = ∂ (xnear − xobstacle )

+ ρ
xrand − xnear
pn ≤ p0
. (18)
∂xnear kxrand − xnear k




0 pn > p 0

In this way, under the action of the component of repul-

sive force, the points in the random tree seek and grow,
in a direction leading away from obstacles in the free Figure 7. Path-planning process based on the improved APF
space, so that the robot can escape from the local mini- method.
mum and reach its objective.

the region of oscillation. When the robot escapes from the

3.3 Implementation of path planning of the robot based local minimum, it returns to path planning using the APF
on the improved APF method method. The path-planning process of the robot is illustrated
The main idea of path planning of the robot using the im- in Fig. 7.
proved APF method is summarized as follows: firstly, the The specific implementation steps for path planning of the
APF method is used for path planning in the workspace of robot based on the APF method are presented as follows:
a robot. When the motion path falls to a local minimum or
the objective point is unreachable, the RRT algorithm is used Step 1. The APF method is used to establish a virtual po-
to optimize the local optimal solution and forces it out of tential field in the workspace. The attractive, repulsive,

Mech. Sci., 14, 87–97, 2023 https://doi.org/10.5194/ms-14-87-2023

Q. Jiang et al.: Path planning for a 5 DOF robot 93

and resultant forces of the potential field are calculated,

respectively, and path planning is conducted.

Step 2. The robot moves to its objective under the guidance

of forces in the potential field. If the objective is un-
reachable, or the algorithm falls into a local minimum
during the motion of the robot, Step 3 is performed. Oth-
erwise Step 4 is executed.

Step 3. The RRT algorithm is utilized to detect and search

for obstacles in the space. By taking the minimum point
as a root node, a search structure in the form of a tree
is generated until the robot escapes from the region of
oscillation.

Step 4. The robot returns to the potential field and continues Figure 8. Schematic diagram of environmental obstacles.
to move towards its objective.

Step 5. Check whether the robot has reached its objective; if

it has not, the algorithm returns to Step 2, and if it has,
the task of path planning ends.

The main characteristics of the improved APF algorithm

are shown as follows:

1. The proposed algorithm is simple and implemented in

a small search space. By improving the traditional po-
tential field function and searching for the path which
escapes only near the local minimum using the RRT al-
gorithm, the optimal path can be rapidly found. In this
way, the time taken to pass an obstacle is reduced, the
probability of convergence of the algorithm is increased,
and the planning time is shortened.

2. The algorithm is highly robust. Compared with algo-

rithms such as neural networks and genetic algorithms,
the APF function can avoid the singularity of the robot Figure 9. Projection trajectory of RRT in XOY plane of robot.
and improve the smoothness of the path.

4 Experiment and verification of the algorithm By using the RRT algorithm, the path of the mechanical
arm moving from initial point Q (50 cm, 350 cm) to the end
The proposed path-planning method of robots based on the point Z (400 cm, 120 cm) is simulated using MATLAB™
improved APF method is able to plan a realizable, collision- (Fig. 9). The parameters of the improved APF method are
free motion path that does not fall into the local optimum generally obtained according to the empirical method, sim-
according to the positions of the initial point, objective point, ulation result, and experimental environment. The parameter
and obstacles. It is applied to the bending operation of a values are set as the step length ρ = 0.5, the scale factor of
bending robot. attractive force ka =1, and the repulsive force of attractive
force kr = 50. The simulation results demonstrate that the
motion path planned by the RRT algorithm contains turn-
4.1 Simulation analysis of the algorithm
ing points, thus increasing the optimization path, making the
In the workspace of the bending robot, by taking the position moving process unsmoothed, and requiring more time for
where the X axis of the mechanical arm moves to the far left computation.
as the origin, three obstacles of different shapes are estab- In the same workspace, the improved APF method is sim-
lished. Their boundary coordinates can be seen in Fig. 8. The ulated. The APF is constructed for the obstacles in the above
schematic diagram of the mechanical arm moving from the space, and the potential field of attractive force at the ob-
initial point Q (50 cm, 350 cm) to the end-point Z (400 cm, jective point and potential field of repulsive force generated
120 cm) is also illustrated in Fig. 8. by obstacles are illustrated in Fig. 10. The resultant potential

https://doi.org/10.5194/ms-14-87-2023 Mech. Sci., 14, 87–97, 2023

94 Q. Jiang et al.: Path planning for a 5 DOF robot

Figure 10. Potential field of the target point and obstacle point.

Figure 12. Projection trajectory of improved APF in XOY plane of

robot.

Table 1. Performance comparison of path-planning algorithms in

the simulation environment.

Path-planning algorithms RRT APF Improved APF

Path length (cm) 565.6 572.8 536.2
Figure 11. Composite potential field of obstacle and target point. Completion time (s) 1.6 2.1 1.5
Accuracy rate (%) 86.31 61.54 98.26

field of the obstacles and the objective point in the motion

space is displayed in Fig. 11. actual working environment to assess the effectiveness of ob-
The trajectory of the robot, from the initial point to the end stacle avoidance of the robot and the optimality of the path.
point, is shown in Fig. 12; the path generated by the improved In the experiment, the position parameters of the
APF algorithm meets the requirements of obstacle avoidance obstacles are established: two cartons, whose size is
and is smooth, avoiding local minima. 400 mm × 300 mm × 280 mm (length × width × height), are
Table 1 lists the length of the optimal path, completion selected as obstacles at central coordinates of (−100 mm,
time, and accuracy of obtaining the optimal path by perform- −900 mm, −260 mm) and (825 mm, 100 mm, −260 mm).
ing 20 simulations of path planning using the RRT algorithm, The actual scene of the robot obstacle-avoidance experiment
APF algorithm, and improved APF algorithm. is shown in Fig. 13.
Therefore, integrating the RRT algorithm into the im- Before the system is started, the mechanical arm of the
proved APF can overcome the shortcomings of being unable robot is situated at an initial position with coordinates (0 mm,
to escape the trap of a local minimum. When the motion plan- 0 mm, 0 mm). The coordinates of the objective point of mo-
ning of the algorithm falls to a local minimum, the local ran- tion input in the system are (525 mm, 240 mm, −240 mm).
dom tree generated by the RRT algorithm guides the robot to The system is then run, and the mechanical arm moves from
continue the process of moving towards the objective point the initial point to the objective. During the motion, the me-
to obtain the optimal planning path. At the same time, the chanical arm avoids obstacles according to the automati-
problems of an unsmoothed path planning and the long plan- cally planned shortest obstacle-avoidance path. The obstacle-
ning time needed in the RRT algorithm are optimized, thus avoidance process of the robot arm to the end point is demon-
accelerating the overall path-planning process. strated in Fig. 13.
The RRT algorithm, APF algorithm, and improved APF
are used to conduct 20 path-planning experiments for obsta-
4.2 Obstacle-avoidance path-planning experiment of cle avoidance during bending, respectively. The results of the
the robot length of the optimal path, completion time, and the accuracy
of the resulting optimal path are listed in Table 2.
To verify the obstacle-avoidance planning ability of the robot The experimental results illustrate that in the presence of
based on the improved APF method, obstacles are set in an obstacles in the potential path of the robot, the improved APF

Mech. Sci., 14, 87–97, 2023 https://doi.org/10.5194/ms-14-87-2023

Q. Jiang et al.: Path planning for a 5 DOF robot 95

Figure 13. The obstacle-avoidance process of the robot arm: (a) the starting state, (b) the obstacle-avoidance state, and (c) the ending state.

Table 2. Performance comparison of path-planning algorithms in meet the technical requirements expected of such a bending
the actual obstacle of the bending robot. robot.
In subsequent studies, dynamic obstacles should be added
Path-planning algorithms RRT APF Improved APF to the existing environment to increase the complexity of the
Path length (mm) 968.3 1019.2 843.9 path-planning environment and to verify whether the algo-
Completion time (s) 3.3 4.9 3.1 rithm can maintain a high accuracy in the environment with
Accuracy rate (%) 83.76 58.61 96.54 dynamic obstacles. Therefore, our next work is to improve
the performance of the proposed method by building an ex-
perimental verification platform.
method can effectively plan an optimal obstacle-avoidance
path to make the robot avoid obstacles in the shortest time.
Data availability. All the data used in this paper can be obtained
The mechanical arm moves smoothly and continuously with-
by request from the corresponding author.
out halting, thus meeting the technical requirements imposed
upon such a bending robot.
Author contributions. QJ conceptualized the study and reviewed
5 Conclusions and edited the paper. KC wrote the original draft of the paper. FX
was responsible for data curation and validation.
A strategy of path planning for a 5 DOF bending robot is
conducted in an environment with obstacles, and the equa-
tions of motion tracks of the mechanical arm are determined. Competing interests. The contact author has declared that none
For the obstacle-avoidance problem, the improved APF algo- of the authors has any competing interests.
rithm, in combination with the idea of a random tree method,
is proposed, and the implementation process of improved
APF is presented. In view of the existence of local min- Disclaimer. Publisher’s note: Copernicus Publications remains
neutral with regard to jurisdictional claims in published maps and
ima in the APF method, based on the random extension
institutional affiliations.
and search characteristics of the RRT algorithm, the robot
can escape from the local minimum, which shortened the
time of path planning and ensured smoothness of motion. Fi-
Financial support. This project is supported by the National
nally, the feasibility of path planning based on the improved Natural Science Foundation of China (grant no. 52175100) and
APF method is verified through MATLAB simulation and the Natural Science Foundation of Jiangsu Province (grant no.
obstacle-avoidance experiments for path planning of the me- BK20201379).
chanical arm. The experimental results illustrate that the pro-
posed path-planning method can find the optimal path and

https://doi.org/10.5194/ms-14-87-2023 Mech. Sci., 14, 87–97, 2023

96 Q. Jiang et al.: Path planning for a 5 DOF robot

Review statement. This paper was edited by Daniel Condurache Kim, J. J. and Lee, J. J.: Trajectory optimization with
and reviewed by two anonymous referees. particle swarm optimization for manipulator motion
planning, IEEE Trans. Ind. Inform., 11, 620–631,
https://doi.org/10.1109/TII.2015.2416435, 2015.
References Korayem, M. H., Hoshiar, A. K., and Nazarahari, M.: A hybrid co-
evolutionary genetic algorithm for multiple nanoparticle assem-
Ahmed, A., Maged, A., Soliman, A., El-Hussieny, H., and bly task path planning, Int. J. Adv. Manuf. Tech., 87, 3527–3543,
Magdy, M.: Space deformation based path planning https://doi.org/10.1007/s00170-016-8683-4, 2016.
for Mobile Robots, ISA Transactions, 126, 666–678, Kovács, B., Szayer, G., Tajti, F., Burdelis, M., and Korondi, P.: A
https://doi.org/10.1016/j.isatra.2021.08.019, 2022. novel potential field method for path planning of mobile robots
Akbaripour, H. and Masehian, E.: Semi-lazy probabilistic roadmap: by adapting animal motion attributes, Robot. Auto. Syst., 82, 24–
a parameter-tuned, resilient and robust path planning method for 34, https://doi.org/10.1016/j.robot.2016.04.007, 2016.
manipulator robots, Int. J. Adv. Manuf. Tech., 89, 1401–1430, Lai, T. C., Xiao, S. R., Aoyama, H., and Wong, C. C.: Path
https://doi.org/10.1007/s00170-016-9074-6, 2017. planning and obstacle avoidance approaches for robot arm,
Cheng, Z., Zhang, Z., Li, J., and Jiang, T.: Mobile robot path plan- in: 2017 56th Annual Conference of the Society of Instru-
ning based on improved artificial potential field, Comput. Eng. ment and Control Engineers of Japan (SICE), 334–337, IEEE,
Appl., 55, 29–34, https://doi.org/10.3778/j.issn.1002-8331.1904- https://doi.org/10.23919/SICE.2017.8105619, 2017.
0472, 2019. Li, C., Huang, X., Ding, J., Song, K., and Lu, S.: Global path
Cong, M., Zhao, S., Liu, D., and Du, Yu.: Handoff location selecting planning based on a bidirectional alternating search A* algo-
and path planning for dual-arm robot, J. Huazhong Univ. Sci., 48, rithm for mobile robots, Comput. Indust. Eng., 168, 108123,
1–6, https://doi.org/10.13245/j.hust.200901, 2020. https://doi.org/10.1016/j.cie.2022.108123, 2022.
Contreras-Cruz, M. A., Ayala-Ramirez, V., and Hernandez- Li, H., Shi, A., and Dai, Z.: A trajectory planning method for
Belmonte, U. H.: Mobile robot path planning using artificial bee sprawling robot inspired by a trotting animal, J. Mech. Sci. Tech-
colony and evolutionary programming, Appl. Soft Comput., 30, nol., 31, 327–334, https://doi.org/10.1007/s12206-016-1235-x,
319–328, https://doi.org/10.1016/j.asoc.2015.01.067, 2015. 2017.
Das, S. D., Bain, V., and Rakshit, P.: Energy optimized robot arm Li, Z., Ma, H., Zhang, X., and Fei, Q.: Path planning of
path planning using differential evolution in dynamic environ- the dual-arm robot based on VT-RRT algorithm, 2019 Chi-
ment, in: 2018 Second International Conference on Intelligent nese Control Conference (CCC), IEEE, 2019, 4359–4364,
Computing and Control Systems (ICICCS), 1267–1272, IEEE, https://doi.org/10.23919/ChiCC.2019.8866388, 2019.
https://doi.org/10.1109/ICCONS.2018.8663106, 2018. Nishi, T. and Mori, Y.: Energy efficient motion planning of dual-
Fan, X., Guo, Y., Liu, H., Wei, B., and Lyu, W.: Improved artifi- armed robots with pickup point determination for transporta-
cial potential field method applied for AUV path planning, Math. tion tasks, in: 2018 IEEE International Conference on Industrial
Prob. Eng., 2020, 1–21, https://doi.org/10.1155/2020/6523158, Engineering and Engineering Management (IEEM), 1401–1405,
2020. IEEE, https://doi.org/10.1109/IEEM.2018.8607814, 2018.
Filipenko, M. and Afanasyev, I.: Comparison of various slam sys- Orozco-Rosas, U., Montiel, O., and Sepúlveda, R.: Mobile
tems for mobile robot in an indoor environment, in: 2018 Inter- robot path planning using membrane evolutionary artifi-
national Conference on Intelligent Systems (IS), 400–407, IEEE, cial potential field, Appl. Soft Comput., 77, 236–251,
https://doi.org/10.1109/IS.2018.8710464, 2018. https://doi.org/10.1016/j.asoc.2019.01.036, 2019.
Gala, D. and Sun, L.: Moving Sound Source Localization Rasekhipour, Y., Khajepour, A., Chen, S. K., and Litkouhi, B.: A
and Tracking Using a Self Rotating Bi-Microphone Array, potential field-based model predictive path-planning controller
in: 2019 ASME Dynamic Systems and Control Confer- for autonomous road vehicles, IEEE T. Intell. Transp., 18, 1255–
ence (DSCC), Park City, Utah, USA, Vol. 59148, https: 1267, https://doi.org/10.1109/TITS.2016.2604240, 2016.
//www.researchgate.net/publication/344722626_MOVING_ Ratiu, M. and Prichici, M. A.: Industrial robot tra-
SOUND_SOURCE_LOCALIZATION_AND_TRACKING_ jectory optimization-a review, in: MATEC web
USING_A_SELF_ROTATING_BI-MICROPHONE_ARRAY of conferences, EDP Sciences, 126, 02005,
(last access: 2 October 2022), 2019. https://doi.org/10.1051/matecconf/201712602005, 2017.
Gala, D., Lindsay, N., and Sun, L.: Three-dimensional sound Shao, J., Luo, D., Xu, Y., and Duan, H.: Cooperative path
source localization for unmanned ground vehicles with a planning for multiple robots with motion constraints in
self-rotational two-microphone array, in: Proceedings of the obstacle-strewn environment, IEEE Access, 7, 132286–132301,
5th International Conference of Control, Dynamic Sys- https://doi.org/10.1109/ACCESS.2019.2939181, 2019.
tems and Robotics, Niagara Falls, ON, Canada, 104, 1–11, Stentz, A.: Optimal and efficient path planning for partially
https://doi.org/10.11159/cdsr18.104, 2018. known environments, Intelligent Unmanned Ground Vehicles:
Han, X. G., Yin, M., Liu, X. G., et al.: Solution of Inverse Autonomous Navigation Research at Carnegie Mellon, 203–220,
Kinematics and Motion Trajectory Simulation for 6R Robot, https://doi.org/10.1007/978-1-4615-6325-9_11, 1997.
J. Sichuan Univ. (Engineering Science Edition), 47, 185–190, Sun, W., Lv, Y., Tang, H., and Xue, M.: Mobile robot
https://doi.org/10.15961/j.jsuese.2015.06.026, 2015. path planning based on an improved A* algorithm,
Jun, L., Zhibing, S., and Suzhou, W.: Robot trajectory J. Hunan Univ. (Natural Sciences), 44, 94–101,
planning based on spline function and improved ge- https://doi.org/10.16339/j.cnki.hdxbzkb.2017.04.013, 2017.
netic algorithm, Manu. Technol. Machine Tool, 7, 91–95,
https://doi.org/10.19287/j.cnki.1005-2402.2017.07.016, 2017.

Mech. Sci., 14, 87–97, 2023 https://doi.org/10.5194/ms-14-87-2023

Q. Jiang et al.: Path planning for a 5 DOF robot 97

Wang, H., Hao, C., Zhang, P., Zhang, M., Yin, P., and Zhang, Yu, J., Wu, P., Yu, N., Zuo, G., and Zhang, Y.: Research
Y.: Path planning of mobile robots based on A* algorithm and and implementation of robot arm task imitation system
artificial potential field algorithm, China Mechan. Eng., 30, based on RNN, in: 2017 IEEE International Conference
2489, http://www.cmemo.org.cn/EN/Y2019/V30/I20/2489 (last on Robotics and Biomimetics (ROBIO), 2484–2489, IEEE,
access: 21 September 2022), 2019. https://doi.org/10.1109/ROBIO.2017.8324793, 2017.
Wang, K. and Wan, H.: Trajectory planning of robot arm, Manufac- Zhou, X., Wang, X., Xie, Z., Li, F., and Gu, X.: Online ob-
turing Automation, 37, 8–11, 2015. stacle avoidance path planning and application for arc weld-
Wang, X. Y., Yang, L., Zhang, Y., and Meng, S.: Robot path ing robot, Robot. Comput.-Integrated Manufact., 78, 102413,
planning based on improved ant colony algorithm with po- https://doi.org/10.1016/j.rcim.2022.102413, 2022.
tential field heuristic, Control and Dec., 33, 1775–1781, Zheng, K., Hu, Y., and Wu, B.: Trajectory planning of multi-degree-
https://doi.org/10.13195/j.kzyjc.2017.0639, 2018. of-freedom robot with coupling effect, J. Mech. Sci. Technol., 33,
Wan, F., Zhou, F., Yin, L., Wang, Y., Chen, K., and Shen, 413–421, https://doi.org/10.1007/s12206-018-1241-2, 2019.
D.: Global path planning algorithm of mobile robot Zhuang, L., Hongbin, M., Xiaofei, Z., and Qing, F.: Path
based on electric potential field, Robot, 41, 742–750, planning of the dual-arm robot based on VT-RRT al-
https://doi.org/10.13973/j.cnki.robot.180687, 2019. gorithm, 2019 Chinese Control Conference (CCC),
Guangzhou, China, 27–30 July 2019, IEEE, 4359–4364,
https://doi.org/10.23919/ChiCC.2019.8866388, 2019.

https://doi.org/10.5194/ms-14-87-2023 Mech. Sci., 14, 87–97, 2023