1

Modified RRT* for Path Planning in Autonomous

Driving
Sugirtha Ta , Pranav Sb , Nitin Benjamin Dasiahb , Sridevi Mb
a Department of CSE, Indian Institute of Information Technology, Kottayam, India
b Department of CSE, National Institute of Technology, Tiruchirappalli-620015, India

Abstract—Essential tasks in autonomous driving includes en- The path planning task is accomplished as a sequence
arXiv:2402.12129v1 [cs.RO] 19 Feb 2024

vironment perception, detection and tracking, path planning and of four subtasks namely (i) Global path planning (ii)
action control. This paper focus on path planning, which is one behavioural planner (iii) local path planning (iv) local feedback
of the challenging task as it needs to find optimal path in highly
complex and dynamic environments. Usually, a driving scenario control.
has large num- ber of obstacles in their route. In this paper, Global path planning refers to finding the collision free
we propose a two-stage path planning algorithm named Angle- optimal path from starting state to the destination state. Local
based Directed Rapidly exploring Random Trees (AD- RRT*) to path planning refers to planning the intermediate local move
address the problem of optimal path in complex environment. by creating the nodes within the coordinate space.
The proposed algorithm uses A* algorithm for global path
planning and modifies RRT* to bound the samples using angle. Various algorithms were proposed over past four decades
The efficiency of the proposed algorithm is evaluated through to solve path-planning problem in static and dynamic en-
experiments in different scenarios based on the location and vironment which includes grid based [37], [38], Artificial
number of obstacles. The proposed algorithm showed higher rate Potential Field (APC) [39], [40], [41], evolutionary methods
of convergence with reduced time and less number of nodes than [42], [43], [44], [45], geometric and neural networks [46], [47].
the base RRT* algorithm.
An extensive survey that describes the merits and demerits of
these path planning algorithms can be found in [48], [49], [50].
I. I NTRODUCTION Though, there were several techniques employed for the path
Autonomous driving has emerged as a hot research topic
in recent decades. Hence, many researchers and automotive
companies spend billions of dollars in developing various
modules to make Level-5 self driving cars. Deep learning
has achieved state-of-the-art results in both semantic and
geometric tasks in autonomous driving which includes object
detection [1], [2], [3], [4], semantic segmentation [5], [6], [7],
[8], depth prediction [9], [10], [11], [12], adverse weather
detection [13], [14], [15], [16], moving object detection [17],
[18], [19], [20], SLAM [21], [22], [23], multi-task learning
[24], [25], [26], [27] and sensor fusion [28], [29], [30]. The
preliminary task is object detection and tracking, and the next Fig. 1: Tree expansion process of RRT* [51]
crucial task in a self-driving car is to find the optimal path
from source to destination point. Path planning helps the ego- planning, the following issues were identified:
vehicle to navigate through the environment by considering • Slow convergence to optimal solution
the surrounding obstacles. • Longer paths from source to destination
Path planning algorithms have widespread applications in • High memory requirement by creating large number of
autonomous driving [31], planetary and space missions [32], nodes
Unmanned Aerial Vehicles (UAVs) [33] , computerized robotic • Not suitable for complex dynamic environment
surgery [34], , artificial intelligence for video games [35], and To address these issues, this paper proposes a variant of RRT*.
molecular biology [36]. The major contributions of this paper is given as follows :
Path planning refers to generation of collision free trajectory
• Proposed an angle based detector in front of the ego-
from an initial state to destination state and a velocity profile
vehicle to quickly find optimal path
to the controller by considering the following factors :
• Produce comparatively shorter path than base RRT*
• State of the self driving car which includes velocity,
• Consumes less memory
inclination etc., • Deals with complex environment
• Surrounding environment of the ego-vehicle which in-
The remaining section of the paper is organized as follows
cludes static and dynamic obstacles, drivable spaces
: Section 2 describes the detailed explanation of various
• Traffic laws and allowed maneuvers
existing path planning algorithms. Section 3 explains the
 2

proposed Angle-based Directed RRT* in detail manner. The in whole configuration space. Also, the space complexity is
experimental results and performance analysis of the proposed high as the number of nodes in the tree is high.
method is provided in Section 4. Finally, Section 5 concludes To address the above drawbacks, a new variant of RRT*,
the paper. named RRT*-smart [57] was proposed. It uses intelligent
sampling and path optimization for faster convergence of
II. R ELATED W ORK optimal path. RRT*-smart finds the initial path similar to
This section explains the local path planning methods RRT* and optimizes the path by performing intelligent
exists in the literature. We classify them into two categories sampling at regular intervals using biasing ratio. RRT*-smart
namely, (i) Potential field based path planning methods and converges faster than RRT*, but adjusting the biasing ratio
(ii) Graph based path planning methods. causes computational overhead. RRT*-Adjustable Bounds
Conventional path planning approaches use Potential (RRT*-AB) address the issues of RRT*-smart such as dense
Field (PF) [52] which includes attractive and repulsive tree and high memory requirements. RRT*-AB finds optimal
potential fields. The vehicle is attracted towards the goal path by exploring only favorable regions and focusing
by the attractive field and repelled from obstacles by the on features like (i) connectivity (ii) intelligent bounded
repulsive field. The limitation of conventional Potential sampling and (iii) path optimization. In order to overcome the
Field Method (PFM) is that it gets stuck at local minima. limitations of RRT and RRT*, we propose a two-stage path
To overcome the limitation of conventional PFM, modified planning algorithm for autonomous driving by modifying
Artificial Potential Field (APF) [53] was proposed where base RRT*.
the repulsive field makes use of goal co-ordinates from In this paper, we focus on the path planning task. To ensure
camera, obstacle co-ordinates from radar and autonomous safety, the surrounding vehicles need to follow correct lane,
vehicle co-ordinate from Global positioning system (GPS) overtake other vehicles when it is safe to do so, stop at the
and computes the improved artificial potential field. Obstacle intersections and take U-turns if required and do efficient
dependent Gaussian Potential Field (GPF) proposed in [54] parking in the allotted space. A lane changing maneuver
calculates repulsive potential field by considering the objects scenario of an autonomous vehicle is shown in Fig 3.
information from the sensor within a threshold and uses yaw
angle information and calculate attractive field to find the
path. III. P ROPOSED W ORK
In recent years, Sampling Based Planning (SBP) methods This section explains the proposed two-stage path planning
are probabilistic complete and they are most extensively used algorithm which finds global path and local path to plan the
for finding the optimal path. They perform random sampling maneuvers. The overall pipeline of various tasks involved in
in search space and are computationally efficient in solving autonomous driving is shown in Fig 4. The proposed work
high dimensional real time complex problems. They create focus on the decision making task which finds the optimal
a roadmap of possible trajectories in configuration space path from source to destination points using Algorithm 1.
without explicitly using the obstacle information.
A study and analysis of various SBP techniques can The proposed algorithm utilizes A* algorithm [58] to create
be found in [39]. Probabilistic Road Map (PRM) [56] a global path for deciding which roads to turn. This global
based approach is widely used in highly structured static path will be used by the local path to decide the angles.
environment and suitable for holonomic constraints. But, RRT and RRT* algorithms are generally very inefficient for
they depend on geometry of obstacles and doesn’t ensure real-time path planning. They tend to search large portions
asymptotic optimality. Rapidly exploring Random Tree (RRT) of the map, that is not necessarily useful in case of vehicles.
is one of the fastest path planning algorithms which supports We propose Angle-Based Directed RRT* (AD-RRT*) to
dynamic cluttered environment and can be extended to not only make the algorithm more efficient for vehicles,
non-holonomic constraints. But, it does not ensure asymptotic but to add another layer of algorithm on top of it, so the
optimality. Due to its advantages, various extensions of algorithm can be tuned to required situations with the help
RRT were proposed to improve the efficiency. A major of hyper-parameters. The proposed method uses angle to
breakthrough was witnessed in high dimensional optimal bound the sample. As the time taken to find a path increases,
path planning algorithms with asymptotic optimality after the the angle increases with expansion of the search area. The
introduction of RRT* [51] in 2010. The major advantage of angle uses the global path for the initial angle. If there is
RRT* is the quick exploration of initial path and optimizes an obstacle and the algorithm is unable to find a path, it
it in successive iterations. The asymptotic optimal property expands its search angle, equally on both sides. After a
of RRT* generates near optimal path when the number local path reaches a given distance towards the global path,
of iterations approach infinity makes it very expedient for the vehicle will move towards it, and updation of the point
real time applications. The tree expansion process of RRT* from which the angle is calculated, reducing the search area
is shown in Fig 1. Nearest neighbour search and rewiring back to 0°. For most cases, it will force the vehicle to go
operations are shown in Fig 2. However, RRT* never produce towards the goal path reaching the final state. This method of
the optimal path in finite time and the convergence rate is bounding the search area, will not always produce an optimal
slow. This happens because it explores and samples the nodes path and in worst case scenarios take more time to search
 3

7 7
3 3
2 2

3 3
5 5 1 5 5 1
8
1 8 1
3 3
Z near Z near
3 4 3 3
4 9 Z new 4 9 Z new
Z start 0 10 Z start 0 10
5 1 5

12 6 12 6
3 3

(a) (b)

7 7
3 3
2 2

3 3
5 5 1 5 5 1
8
1 8 1
3 3
Z near Z near
3 4 3 3
4 9 Z new 4 9 Z new
Z start 0 10 0 10
5 1
1
12 6 12
3 3 6

(c) (d)

Fig. 2: Rewiring Process of RRT* [51]. (a) Finding near vertices (b) Selection of best parent (c) cost check (d) rewiring with
minimum cost

Fig. 3: Lane changing maneuver [55]

Environment Decision
Control System
Perception Making

Object Global path

detection planning Actuator
Sensor input
(A* algorithm)

Object
tracking
Local path
planning
(Proposed
Angle-based
Directed RRT*)

Fig. 4: General pipeline of autonomous driving tasks [focus on Decision making task]

for path. Therefore, the proposed algorithm is best used for denotes the obstacle free region. Zsrc and Zdestn denotes the
vehicles and needs more optimizations to work efficiently. starting and destination points respectively. u : [0 : s] denotes
It is required to use visibility regions to search in shorter the path connecting the states z1 and z2 such that u(0) = z1
distances than longer distances when there are more crowd. and u(s) = z2 , where s is the positive scalar length of the
This allows for searching through narrow areas if needed. path. Σf denotes the set of paths in Zfree that are collision
free and σf ∈ Zfree denotes end to end feasible path. CRegion
Notation : denotes a region with connected nodes near to the starting
Let Z ⊂ Rn , n ∈ N represents the configuration space and node Zsrc .
Zobs represents the obstacle region and Zfree = Z/Zobs
 4

Algorithm 1 Angle based Directed RRT*

1: T ← InitializeTree ()
2: T ← InsertNode (ϕ, zsrc , T )
3: Angle ← 0°
4: ALength ← collectf rompreviouslayervisibility
5: CRegion ← ConnectivityRegion (zsrc , ALength , Aangle , T )
6: for i ← 0 to N do
7: zrand ← Bounded Sample (i, CRegion )
8: znearest ← Nearest (T, zrand )
9: (znew , Unew ) ← Steer (znearest , zrand )
10: if CollisionCheck (znew ) then
11: znear ← Near (T, znew , |V |)
12: zmin ← ChooseParent (znear , znearest , znew )
13: T ← Insert Node (zmin , znew , T )
14: T ← Rewire (T, znear , zmin , znew )
15: if P athF ound(T ) then
16: CRegion ← ConnectivityRegion (zinit , zdestn , T )
17: else if T imetakenhigh ()
18: CRegion ← ConnectivityRegion (zinit , zgoal , T )
19: angle ← angle + increase
20: CRegion ← ConnectivityRegion (zsrc , ALength , Aangle , T )
21: T ← Prune Path (T )
22: return
23: end if
24: end if
25: end for

TABLE I: Comparison of base RRT* with Proposed Angle-based Directed RRT*

Base RRT* Angle based Directed RRT* (AD-RRT*)
No. of obstacles
Time taken Time taken
No. of nodes Total path cost Average path cost No. of nodes Total path cost Average path cost
(s) (s)
50 8174 21.28 41203.74 934.01 6904 20.7 39556.83 791.13
68 9070 30.05 39226.16 784.52 8534 31.42 38502.22 770.04
70 11803 78.74 46700.4 934.01 8656 67.18 43174.01 863.48
80 15353 59.5 44426.36 888.52 10461 47.55 41802.47 836.04
85 13198 56.48 44419.73 888.39 11834 56.27 41530.99 830.62
100 16785 99.72 47707.62 954.15 12567 88.23 44661.59 893.23

The proposed algorithm finds a feasible path σf ∈ Σf Steer : The procedure Steer (znear , zrand ) provides a control
from σf (0) = zsrc to σf (s) = zdestn by building a Tree input u[0, T ] that drives the system from x(0) = zrand to
T = (V, E), that has the vertices (V ) sampled from obstacle x(T ) = znear along the path x : [0, T ] → X giving znew
free region Zfree and (E), the edges that connect these vertices. at a distance ∆q from znear towards zrand , where ∆q is the
incremental distance.
The major procedures of the algorithm uses the following Collision check : This procedure determines if a path
task : u[0, T ] exists in the obstacle free region Zf ree for all t = 0
to t = T .
Bounded-sampling : A state Zrand is placed randomly
from obstacle free region Zf ree . Near : This procedure returns the nearby neighbouring
nodes as given by equation (2):
Connectivity Region : A connectivity region CRegion is
k = γ(log(n)/n)(1/d) (2)
identified in search space near to 𭟋src using the area with
length. Choose parent : It selects the best parent 𭟋min which
gives the minimum cost from the source point to 𭟋new .
Nearest : The nearest node from T = (V, E) to zrand is
identified by N earest(T, zrand ) based on the cost computed Insert node : It adds a node 𭟋min to V in the tree
by the distance function as given in equation (1). T = (V, E) and connects it to 𭟋min as its parent.
q
2 2
∆d = (x2 − x1 ) + (y2 − y1 ) (1) Rewire : Rewiring is performed after the parent is selected.
It computes the cost of all nodes through 𭟋new and compared
 5

Scenario 1 Scenario 2

Base RRT*

Proposed
AD - RRT*

(a) (b) (a) (b)

Scenario 3 Scenario 4

Base RRT*

Proposed
AD - RRT*

(a) (b) (a) (b)

Scenario 5 Scenario 6

Base RRT*

Proposed
AD - RRT*

(a) (b) (a) (b)

Fig. 5: Visual results of Path planning for Scenarios S1, S2, S3, S4, S5 and S6. (a) and (b) under each scenario denotes the
intermediate and final step in path planning.

(a) (b)

(c) (d)

Fig. 6: Visual results of proposed method on CARLA Simulator (a) Plot of global path planning, Local path planning - (b)
Initial step (c) Intermediate step and (d) Final step where the ego-vehicle takes the right turn

with their previous costs. If the cost is minimum for any sampling and no path is found in the first scan and the time
node, then 𭟋new will be made as its new parent. exceeds the limit, then the angle of CRegion is increased as
mentioned in Steps 15-19 of the Algorithm 1.
Prunepath : It returns the nodes that are connectable from
source to destination point in the obstacle free region. A tree
T is initialized with 𭟋src as the root node. Bounded sampling IV. E XPERIMENTAL RESULTS AND A NALYSIS
randomly selects nodes restricted by an angle from CRegion This section demonstrates the simulation results of proposed
and the tree is populated with these nodes. CRegion is defined Angle-Based Directed RRT* with CARLA simulator [59].
by expansion distance scale in a search space between 𭟋src The experiments were carried out with different scenario
and 𭟋destn using equation (3): based on the density (cluster) of the obstacle nodes. The
Dscale = E/m (3) proposed approach is implemented using Python in an Intel
i7, NVIDIA 1080 GTX GPU.
where, E – size of the environment map
m – expansion factor The proposed algorithm is experimented on six different
When a single scan of CRegion is completed using bounded scenarios i.e. ( S1, S2, S3, S4, S5 and S6) which represents
 6

the distribution of obstacle nodes (N). The scenarios are R EFERENCES

explained as follows : [1] M. Klingner, S. Borse, V. R. Kumar, B. Rezaei, V. Narayanan, S. Yoga-
mani, F. Porikli, X3kd: Knowledge distillation across modalities, tasks
and stages for multi-camera 3d object detection, in: Proceedings of the
1) S1 – Obstacles densed near starting point with N=50 IEEE/CVF Conference on Computer Vision and Pattern Recognition,
2) S2 - Sparse distribution of obstacles throughout the path 2023, pp. 13343–13353.
from starting to destination point with N=50 [2] S. Mohapatra, S. Yogamani, H. Gotzig, S. Milz, P. Mader, Bevdetnet:
bird’s eye view lidar point cloud based real-time 3d object detection for
3) S3 - Sparse distribution of obstacles throughout the path autonomous driving, in: 2021 IEEE International Intelligent Transporta-
from starting to destination point with N=68 tion Systems Conference (ITSC), IEEE, 2021, pp. 2809–2815.
4) S4 – Entire region equally densed with obstacles N=70 [3] H. Rashed, E. Mohamed, G. Sistu, V. R. Kumar, C. Eising, A. El-
Sallab, S. Yogamani, Fisheyeyolo: Object detection on fisheye cameras
5) S5 - Obstacles densed in middle region with N=80 for autonomous driving, in: Proceedings of the Machine Learning for
6) S6 - Obstacles densed near starting, middle and destina- Autonomous Driving NeurIPS 2020 Virtual Workshop, Virtual, Vol. 11,
tion region with N=85 2020.
[4] S. Thayalan, S. Muthukumarasamy, K. Santhakumar, K. B. Ravi, H. Liu,
Fig 5 shows the path planning results of base RRT* and T. Gauthier, S. Yogamani, Monocular three-dimensional object detection
the proposed Angle-Based Directed RRT* under different using data augmentation and self-supervised learning in autonomous
driving, Journal of Electronic Imaging 32 (1) (2022) 011004. doi:
scenarios. The global path planning A* algorithm finds the 10.1117/1.JEI.32.1.011004.
path between Zsrc and Zdestn . The local path planning is [5] A. Briot, P. Viswanath, S. Yogamani, Analysis of efficient cnn design
performed by the proposed Angle-Based Directed RRT*. techniques for semantic segmentation, in: Proceedings of the IEEE
Conference on Computer Vision and Pattern Recognition Workshops,
From the Figure, it is clear that the base RRT* expands the 2018, pp. 663–672.
tree in entire search space and thus results in generating [6] S. Chennupati, G. Sistu, S. Yogamani, S. Rawashdeh, Auxnet: Auxiliary
dense tree with more number of unnecessary nodes. On the tasks enhanced semantic segmentation for automated driving, arXiv
preprint arXiv:1901.05808.
contrary, the proposed AD-RRT* restricts the search space [7] G. Sistu, I. Leang, S. Yogamani, Real-time joint object detection and
to an angle and finds the optimal path by generating less semantic segmentation network for automated driving, arXiv preprint
number of nodes i.e the convergence rate is high with less arXiv:1901.03912.
[8] M. Siam, S. Valipour, M. Jagersand, N. Ray, S. Yogamani, Convolutional
time. gated recurrent networks for video semantic segmentation in automated
driving, in: 2017 IEEE 20th International Conference on Intelligent
Table I shows the comparison between base RRT* and Transportation Systems (ITSC), IEEE, 2017, pp. 1–7.
[9] V. R. Kumar, S. Milz, C. Witt, M. Simon, K. Amende, J. Petzold,
proposed AD-RRT* for varying number of obstacles. The S. Yogamani, T. Pech, Near-field depth estimation using monocular
proposed method finds optimal path in less time by generating fisheye camera: A semi-supervised learning approach using sparse lidar
less number of nodes. Also, the total path cost and average data, in: CVPR Workshop, Vol. 7, 2018, p. 2.
[10] V. R. Kumar, M. Klingner, S. Yogamani, M. Bach, S. Milz, T. Fin-
path cost are less compared to base RRT*. Fig 6 shows the gscheidt, P. Mäder, Svdistnet: Self-supervised near-field distance estima-
output of CARLA simulator. tion on surround view fisheye cameras, IEEE Transactions on Intelligent
Transportation Systems 23 (8) (2021) 10252–10261.
[11] A. R. Sekkat, Y. Dupuis, V. R. Kumar, H. Rashed, S. Yogamani,
P. Vasseur, P. Honeine, Synwoodscape: Synthetic surround-view fisheye
V. C ONCLUSION camera dataset for autonomous driving, IEEE Robotics and Automation
Letters 7 (3) (2022) 8502–8509.
In this paper, a two-stage path planning algorithm called [12] V. R. Kumar, S. Yogamani, M. Bach, C. Witt, S. Milz, P. Mäder,
Angle-based Directed RRT* is proposed to find the optimal Unrectdepthnet: Self-supervised monocular depth estimation using a
path for autonomous vehicles in complex driving environ- generic framework for handling common camera distortion models,
in: 2020 IEEE/RSJ International Conference on Intelligent Robots and
ments. The proposed method deployed A* algorithm to find Systems (IROS), IEEE, 2020, pp. 8177–8183.
global path and compute the local path in restricted search [13] M. M. Dhananjaya, V. R. Kumar, S. Yogamani, Weather and light
space by modifying RRT*. Analysis of the proposed algo- level classification for autonomous driving: Dataset, baseline and active
learning, in: 2021 IEEE International Intelligent Transportation Systems
rithm is carried out in six different scenarios using CARLA Conference (ITSC), IEEE, 2021, pp. 2816–2821.
simulator and the results show faster convergence with less [14] M. Uricár, J. Ulicny, G. Sistu, H. Rashed, P. Krizek, D. Hurych,
number of nodes than the base RRT* algorithm. In future, the A. Vobecky, S. Yogamani, Desoiling dataset: Restoring soiled areas
on automotive fisheye cameras, in: Proceedings of the IEEE/CVF
proposed algorithm is extended to dynamic environment and International Conference on Computer Vision Workshops, 2019, pp. 0–
the performance is analysed. 0.
[15] M. Uricár, D. Hurych, P. Krizek, et al., Challenges in designing
datasets and validation for autonomous driving, in: Proceedings of the
VI. ACKNOWLEDGEMENT International Conference on Computer Vision Theory and Applications,
2019.
This publication is an outcome of the R&D work undertaken [16] S. Ramachandran, G. Sistu, J. McDonald, S. Yogamani, Woodscape fish-
in the project under TiHAN Faculty Fellowship of NMICPS eye semantic segmentation for autonomous driving–cvpr 2021 omnicv
Technology Innovation Hub on Autonomous Navigation Foun- workshop challenge, arXiv preprint arXiv:2107.08246.
[17] M. Siam, H. Mahgoub, M. Zahran, S. Yogamani, M. Jagersand, A. El-
dation being implemented by Department of Science & Tech- Sallab, Modnet: Moving object detection network with motion and
nology National Mission on Interdisciplinary Cyber-Physical appearance for autonomous driving, arXiv preprint arXiv:1709.04821.
Systems (DST NMICPS) at IIT Hyderabad. [18] H. Rashed, A. El Sallab, S. Yogamani, M. ElHelw, Motion and depth
augmented semantic segmentation for autonomous navigation, in: Pro-
ceedings of the IEEE/CVF Conference on Computer Vision and Pattern
Recognition Workshops, 2019, pp. 0–0.
 7

[19] E. Mohamed, M. Ewaisha, M. Siam, H. Rashed, S. Yogamani, [41] C. L. Goerzen, Z. Kong, B. Mettler, A survey of motion planning
W. Hamdy, M. El-Dakdouky, A. El-Sallab, Monocular instance motion algorithms from the perspective of autonomous uav guidance, Journal
segmentation for autonomous driving: Kitti instancemotseg dataset and of Intelligent and Robotic Systems 57 (2010) 65–100.
multi-task baseline, in: 2021 IEEE Intelligent Vehicles Symposium (IV), [42] J.-H. Liang, C.-H. Lee, Efficient collision-free path-planning of multiple
IEEE, 2021, pp. 114–121. mobile robots system using efficient artificial bee colony algorithm, Adv.
[20] B. R. Kiran, K. Ramakrishnan, Y. S. Kumar, K. Anoop, An improved Eng. Softw. 79 (C) (2015) 47–56. doi:10.1016/j.advengsoft.
connected component labeling by recursive label propagation, in: 2011 2014.09.006.
National Conference on Communications (NCC), IEEE, 2011, pp. 1–5. [43] X. Zhang, Y. Zhao, N. Deng, K. Guo, Dynamic path planning algorithm
[21] N. Tripathi, S. Yogamani, Trained trajectory based automated parking for a mobile robot based on visible space and an improved genetic
system using visual slam on surround view cameras, arXiv preprint algorithm, International Journal of Advanced Robotic Systems 13 (3)
arXiv:2001.02161. (2016) 91. doi:10.5772/63484.
[22] L. Yahiaoui, J. Horgan, B. Deegan, S. Yogamani, C. Hughes, P. Denny, [44] R. Kala, Multi-robot path planning using co-evolutionary genetic pro-
Overview and empirical analysis of isp parameter tuning for visual gramming, Expert Systems with Applications 39 (3) (2012) 3817–3831.
perception in autonomous driving, Journal of Imaging 5 (10) (2019) doi:https://doi.org/10.1016/j.eswa.2011.09.090.
78. [45] W. Zhu, H. Duan, Chaotic predator–prey biogeography-based opti-
[23] L. Yahiaoui, J. Horgan, S. Yogamani, C. Hughes, B. Deegan, Impact mization approach for ucav path planning, Aerospace Science and
analysis and tuning strategies for camera image signal processing Technology 32 (1) (2014) 153–161.
parameters in computer vision, in: Irish Machine Vision and Image [46] X. Wang, Z.-G. Hou, F. Lv, M. Tan, Y. Wang, Mobile robots’ modular
Processing conference (IMVIP), Vol. 2, 2011. navigation controller using spiking neural networks, Neurocomput. 134
[24] I. Leang, G. Sistu, F. Bürger, A. Bursuc, S. Yogamani, Dynamic (2014) 230–238. doi:10.1016/j.neucom.2013.07.055.
task weighting methods for multi-task networks in autonomous driving URL https://doi.org/10.1016/j.neucom.2013.07.055
systems, in: 2020 IEEE 23rd International Conference on Intelligent [47] D. Zhu, W. Li, M. Yan, S. X. Yang, The path planning of auv based
Transportation Systems (ITSC), IEEE, 2020, pp. 1–8. on d-s information fusion map building and bio-inspired neural network
[25] G. Sistu, I. Leang, S. Chennupati, S. Yogamani, C. Hughes, S. Milz, in unknown dynamic environment, International Journal of Advanced
S. Rawashdeh, Neurall: Towards a unified visual perception model for Robotic Systems 11.
automated driving, in: 2019 IEEE Intelligent Transportation Systems [48] M. S. Nosrati, R. Karimi, H. A. Hasanvand, Investigation of the * (star)
Conference (ITSC), IEEE, 2019, pp. 796–803. search algorithms: Characteristics, methods and approaches - ti journals,
[26] H. Rashed, S. Yogamani, A. El-Sallab, P. Krizek, M. El-Helw, Optical World Applied Programming.
flow augmented semantic segmentation networks for automated driving, [49] M. Costa, M. Silva, A survey on path planning algorithms for mobile
arXiv preprint arXiv:1901.07355. robots, 2019, pp. 7–18. doi:10.1109/ICARSC.2019.8733623.
[27] V. R. Kumar, C. Eising, C. Witt, S. Yogamani, Surround-view fisheye [50] I. Noreen, A. Khan, Z. Habib, Optimal path planning using rrt* based
camera perception for automated driving: Overview, survey & chal- approaches: A survey and future directions, International Journal of
lenges, IEEE Transactions on Intelligent Transportation Systems. Advanced Computer Science and Applications 7.
[28] M. Pöpperli, R. Gulagundi, S. Yogamani, S. Milz, Capsule neural [51] S. Karaman, E. Frazzoli, Sampling-based algorithms for optimal motion
network based height classification using low-cost automotive ultrasonic planning, The International Journal of Robotics Research 30 (2011) 846
sensors, in: 2019 IEEE Intelligent Vehicles Symposium (IV), IEEE, – 894.
2019, pp. 661–666. [52] M. A. P. Castañeda, J. Savage, A. Hernández, F. A. Cosı́o, Local
[29] K. Dasgupta, A. Das, S. Das, U. Bhattacharya, S. Yogamani, Spatio- autonomous robot navigation using potential fields, in: X.-J. Jing (Ed.),
contextual deep network-based multimodal pedestrian detection for Motion Planning, IntechOpen, Rijeka, 2008, Ch. 1. doi:10.5772/
autonomous driving, IEEE transactions on intelligent transportation 6022.
systems 23 (9) (2022) 15940–15950. URL https://doi.org/10.5772/6022
[30] C. Eising, J. Horgan, S. Yogamani, Near-field perception for low- [53] F. Bounini, D. Gingras, H. Pollart, D. Gruyer, Modified artificial
speed vehicle automation using surround-view fisheye cameras, IEEE potential field method for online path planning applications, in: 2017
Transactions on Intelligent Transportation Systems 23 (9) (2021) 13976– IEEE Intelligent Vehicles Symposium (IV), 2017, pp. 180–185. doi:
13993. 10.1109/IVS.2017.7995717.
[31] X. Lan, S. D. Cairano, Continuous curvature path planning for semi- [54] J. Cho, D. S. Pae, M.-T. Lim, T.-K. Kang, A real-time obstacle avoidance
autonomous vehicle maneuvers using rrt, 2015 European Control Con- method for autonomous vehicles using an obstacle-dependent gaussian
ference (ECC) (2015) 2360–2365. potential field, Journal of Advanced Transportation.
[32] G. Lau, H. H.-T. Liu, Real-time path planning algorithm for autonomous [55] B. Mashadi, M. Majidi, Global optimal path planning of an autonomous
border patrol: Design, simulation, and experimentation, Journal of vehicle for overtaking a moving obstacle (2014).
Intelligent & Robotic Systems 75 (2014) 517–539. [56] L. Kavraki, P. Svestka, J.-C. Latombe, M. Overmars, Probabilistic
[33] X. Zhang, J. Chen, B. Xin, Path planning for unmanned aerial vehicles in roadmaps for path planning in high-dimensional configuration spaces,
surveillance tasks under wind fields, Journal of Central South University IEEE Transactions on Robotics and Automation 12 (4) (1996) 566–580.
21 (2014) 3079–3091. doi:10.1007/s11771-014-2279-7. doi:10.1109/70.508439.
[34] N. Ahmidi, G. D. Hager, L. Ishii, G. L. Gallia, M. Ishii, Robotic path [57] J. Nasir, F. Islam, U. Malik, Y. Ayaz, O. Hasan, M. Khan, M. S.
planning for surgeon skill evaluation in minimally-invasive sinus surgery, Muhammad, Rrt*-smart: A rapid convergence implementation of rrt*,
in: N. Ayache, H. Delingette, P. Golland, K. Mori (Eds.), Medical International Journal of Advanced Robotic Systems 10 (7) (2013) 299.
Image Computing and Computer-Assisted Intervention – MICCAI 2012, doi:10.5772/56718.
Springer Berlin Heidelberg, Berlin, Heidelberg, 2012, pp. 471–478. [58] S. Sedighi, D.-V. Nguyen, K.-D. Kuhnert, Guided hybrid a-star path
[35] K. Naderi, J. Rajamäki, P. Hämäläinen, Rt-rrt* a real-time path planning planning algorithm for valet parking applications, in: 2019 5th Interna-
algorithm based on rrt, in: Proceedings of the 8th ACM SIGGRAPH tional Conference on Control, Automation and Robotics (ICCAR), 2019,
Conference on Motion in Games, 2015, pp. 113–118. pp. 570–575. doi:10.1109/ICCAR.2019.8813752.
[36] J. Reif, Z. Sun, Nano-robotics motion planning and its applications [59] A. Dosovitskiy, G. Ros, F. Codevilla, A. Lopez, V. Koltun, CARLA:
in nanotechnology and biomolecular computing, in: NSF Design and An open urban driving simulator, in: Proceedings of the 1st Annual
Manufacturing Grantees Conference, 1999, pp. 5–8. Conference on Robot Learning, 2017, pp. 1–16.
[37] P. E. Hart, N. J. Nilsson, B. Raphael, A formal basis for the heuristic
determination of minimum cost paths, IEEE Transactions on Systems
Science and Cybernetics 4 (2) (1968) 100–107. doi:10.1109/
TSSC.1968.300136.
[38] K. Daniel, A. Nash, S. Koenig, A. Felner, Theta*: Any-angle path
planning on grids, J. Artif. Intell. Res. (JAIR) 39. doi:10.1613/
jair.2994.
[39] M. Elbanhawi, M. Simic, Sampling-based robot motion planning: A
review, IEEE Access 2 (2014) 56–77. doi:10.1109/ACCESS.
2014.2302442.
[40] Y. K. Hwang, N. Ahuja, Gross motion planning—a survey, ACM
Comput. Surv. 24 (1992) 219–291.