Robot navigation and target capturing using nature-inspired approaches in a dynamic environment

11/06/2019 ∙ by Devansh Verma, et al. ∙ 34

Path Planning and target searching in a three-dimensional environment is a challenging task in the field of robotics. It is an optimization problem as the path from source to destination has to be optimal. This paper aims to generate a collision-free trajectory in a dynamic environment. The path planning problem has sought to be of extreme importance in the military, search and rescue missions and in life-saving tasks. During its operation, the unmanned air vehicle operates in a hostile environment, and faster replanning is needed to reach the target as optimally as possible. This paper presents a novel approach of hierarchical planning using multiresolution abstract levels for faster replanning. Economic constraints like path length, total path planning time and the number of turns are taken into consideration that mandate the use of cost functions. Experimental results show that the hierarchical version of GSO gives better performance compared to the BBO, IWO and their hierarchical versions.

READ FULL TEXT VIEW PDF
POST COMMENT

Comments

There are no comments yet.

Authors

page 3

page 6

This week in AI

Get the week's most popular data science and artificial intelligence research sent straight to your inbox every Saturday.

1. Introduction

The modern-day robots are made adept in performing several complex tasks by themselves, which usually required human assistance. Some of the recent developments like the Mars Spirit Robot and Self-Driving Cars (Levinson et al., 2011) are an excellent example of today’s technological advancements in the fields of mobile robots. The robots of today are extremely intelligent and are capable of finding their way of completing the task given to them. One of the most crucial functions of mobile robots is to move from one place to another safely without damaging themselves or their corresponding environment.

Robotic navigation is one of the toughest robotic tasks. This problem can be divided into different categories by the following criteria:

  • Type of Environment

    : The type of environment can be broadly classified as 2D and 3D

    (Pandey et al., 2018). The algorithms employed for both of them vary in terms of complexity, approach and applicability. The 2D environment involves two coordinates with the third coordinate is constant, whereas in 3D, the third coordinate is also varying.

  • Nature of Environment: It refers to the environment in which the robot operates, which can be either static or dynamic. Static environment refers to an environment in which the number and the position of obstacles remain constant with respect to time (Masehian and Amin-Naseri, 2006). Due to static nature obstacles, global path planning algorithms can be applied efficiently in order to solve the problem. On the contrary, in a dynamic environment, the environment can change while an agent is deliberating. The speed, position as well as the shape of an obstacle may change with time. Since the same area needs to be explored, again and again, global path planning cannot be applied, so the only choice is local path planning.

  • Planning approach

    : The robot navigation problem can be classified based on knowledge gained at a local and global level. In the case of global path planning, the agent has full prior knowledge about shape, orientation, location and arrangement of the obstacles. Using this information, the path planning algorithm minimizes the cost function in order to reach the target. This approach can only be applied in case of a static environment where obstacles do not move. On the contrary, in local path planning, the robot has little to no prior information about the environment and surroundings. The robot senses the environment within its sensory range and with the help of some meta-heuristic approach plans its next steps while moving in the environment.


Target Searching is a problem of estimation of the location of the target(s) in a given region. This region can be air, water or land. The targets can be mobile as well as stationary and can be of evasive or non-evasive nature

(Albert and Imsland, 2018)

. The robot only has to search for the specific area only once in the case of static targets. On the contrary, for dynamic targets, the same area needs to be explored multiple times as the probability of finding the target is non-zero in that area since the area can be occupied again. For simulation, the target is assumed to be a point object. In practical scenarios, the target can be a person, an animal, wildfire or even a flying aircraft. There are many application such as in search and rescue operations in disastrous areas

(Kantor et al., 2003), robots that can be used for locating mines (Acar et al., 2003), robots used for planetary exploration (Schilling and Jungius, 1996), and in search of remote fires (Marjovi et al., 2009).

Nature has always been a constant source of inspiration for humans (Shukla and Tiwari, 2017). Natural phenomenons have motivated us immensely and enabled us for better innovation. Many laws and theories have been formulated observing different aspects of nature, whether it be plants, animals, birds or whatnot. Nature Inspired Algorithms are algorithms inspired by natural processes regarding biological, geographical, and chemical processes that take place in the natural world. These algorithms have immense potential for solving many real-life problems. A tremendous amount of research has taken place in the past two decades in this field.

2. Related Works

The modern-day robotics is pro efficient in performing several complex tasks in intelligent and automated fashion (Pandey et al., 2018). It is also one of the most difficult as well as a challenging task in a multi-robot system. The implementation of real-life constraints like energy consumption travel cost, path length also plays a significant role in complicating the task even further (Janis and Bade, 2016).

The path planning can be done both in the air as well as in water (Meghjani et al., 2016)

and also in 2D or 3D environment. Several Machine Learning and nature-inspired algorithms like A*, Bat Algorithm

(Wang et al., 2016), PSO (Bilbeisi et al., 2015) are applied too. Many existing algorithms, like Ant Colony Algorithm (Zhang et al., 2017), Invasive Weed Optimization (Mohanty and Parhi, 2014), have been applied in order to tackle this problem. Path planning is purely an optimization problem as it requires optimizing the cost from source to the destination. It is an NP-hard problem and therefore, there is a need for a framework for autonomous robots (Liu et al., 2018) to optimally find their paths. In (Phillips and Likhachev, 2011), the problem has been solved by dividing time into the safe and unsafe interval and acting accordingly. The problem has also been solved using the fuzzy logic approach (Pandey et al., 2014). In (Drake et al., 2018) incremental path planning algorithm for dynamic target has been used that uses past data to update path by sacrificing optimally for reducing computational complexity, and in (Juliá et al., 2012) estimation is used in order to search for the target.One aim of any path planning problem is reaching the target. This target can be anything from a big object to a small object. The prime objective of the navigation problem is aimed at seizing the target. The target can be fixed or mobile. In case of a fixed target, the robot only needs to look for the specific area once. On the contrary, for moving target, the same area needs to be explored multiple times as the probability of finding the target is non-zero at that point as the area can be occupied again. The targets can be located anywhere, be it land or water (Meghjani et al., 2016)

. For detection of a target, specific Gaussian Mixture Models are used in cyber cars

(Premebida and Nunes, 2006). The target may also live for some time and may disappear after that (Jain et al., 2017). In (Zhang et al., 2018), target tracking is done based on some patterns. The target searching problem can be done with or without (Chaudhary and Sinha, 2012) previous information.

In (Dadgar et al., 2016) PSO is used with multiple robots for pathfinding. In (Jensen and Gini, 2018) there has been a comparison between various algorithms for multi-robot discipline. Deep CNN has also been applied in order to solve this problem (Cui et al., 2017). In (Saldana et al., 2015) coordination among the robots is done for detection and tracking of anomalies. In(Hayat and Kausar, 2015) SA algorithm is used to achieve an optimum collision-free path with mobile robots in a static environment containing circular obstacles. The creation of the map is not required in order to solve the path planning problem (Premebida and Nunes, 2006). In (Zhang et al., 2018), multiple robots coordinate among them self treating the densities of robots and targets as properties of the environment.

2.1. Research Gaps

Although much research has been done in the field of robot navigation and targets capturing, it is still not sufficient enough. This paper tries to overcome some of these research gaps through the proposed techniques. Here are some of the research gaps present in this field of robot navigation and target capturing:

  1. The solution obtained by the deterministic algorithm is unfeasible as it is time-consuming. The meta-heuristic algorithms provide sub-optimal paths but in very less time as compared to that of deterministic algorithms.

  2. Not much work has been done in the field of multiresolution path planning using abstract levels. The use of hierarchical levels results in faster replanning.

3. Methodology

3.0.1. Proposed Workflow

The complete project is divided into following modules:

  • Environment Creation: The first step is to create an arena for simulation. It includes designing and modelling the static and dynamic target and obstacles.

  • Algorithm Selection: TIn the next step, the objective is to devise a strategy to generate collision-free paths which involve continuous sensing of the environment and frequent replanning.

  • Path smoothening and refinement: The trajectory generated by the path planning algorithms is generally not smooth and contains several loops and sharp turns. Path smoothening methods such as Catmull- Rom splines are used to generate smooth, realistic paths.

The proposed methodology is shown in Figure 1. It starts with the creation of the arena, which includes adding obstacles and targets. The obstacles and targets are both static as well as dynamic and time-varying. In the path planning phase, the robot senses the environment and broadcast this information to the semi-centralized server, which is employed to achieve the coordination task among the robots in case of a multi-robot system. The received data from the server is used for path planning and moving towards the target. If there is no collision detected, the robot is moved to the required position and checked for target reachability. If a possible collision is detected, the collision avoidance algorithm is triggered, which includes different strategies for dealing with static and dynamic obstacles. After collision avoidance, the path planning system is kicked again, and the process continues.

Figure 1. Proposed methodology

3.0.2. Glowworm Swarm Optimization(GSO)

It is a swarm intelligence based algorithm inspired by a lightning worm called a glowworm. It was proposed by K.N. Krishnand and D. Ghose in the year 2005. GSO has a wide variety of applications like localization, swarm intelligence, clustering, routing and many more.

Glowworm is a small worm that emits light due to a light-emitting compound called luciferin. Each glowworm is capable of adjusting its luciferin content based on environmental factors. The light thus generated is used to lure the prey or attract other worms. It is this behaviour of glowworm is exploited in the GSO algorithm for solving various optimization problems. The worms in the swarm are distributed randomly in search space, having some initial luciferin value. The higher level of luciferin content means a better position of glowworm in the search space. The glowworms then calculate the fitness in the current space using some fitness function and then broadcast this information to all the neighbours within the search radius. The glowworm is attracted to other worm having higher luciferin content and within its search space. These movements which are derived based on the local information provided by the neighbours and selective interaction with the neighbours enables the swarm to divided into disjoint clusters which converges at different points containing the local optima to find multiple solutions. It is similar to other swarming algorithms like ACO and PSO.

The GSO algorithm works in several phases which are responsible for the movement of glowworm in the search space. Figure 2 shows a flowchart of the GSO. It starts by adding a randomly generated population of n glowworms. All the glowworms have same initial luciferin content. The algorithm starts with luciferin-updation phase followed by the movement phase. The phases in GSO can be explained as follows: -

Glow-worm Distribution Phase: This is the first step of the algorithm. The swarm of glowworm is randomly dispersed in the search space each glowworm having an initial luciferin value of and a search radius .

Luciferin Updatation Phase: This phase changes the luciferin content of the glowworms based on the function value at the position of the glowworm. The luciferin content is enhanced based on the objective function value. Simultaneously the value is decreased to simulate the degradation of luciferin with time. The luciferin update phase can be stated as follows:

(1)

where and are decay constant and luciferin enhancement constants respectively. denotes the luciferin content of glowworm at iteration. denotes the objective function.

Movement Phase: In this phase the glowworm decides to move towards other neighbouring glowworms based on probabilistic mechanism. The glowworm is attracted to other glowworm which is within its range and have luciferin content greater than its own. The equation governing the probability to move towards neighbour glowworm is as follows:

(2)

where is an ordered set of neighbouring glow-wworm, according to the below formula:

(3)

where denotes the distance between and glowworm and the search radius of glowworm is denoted by . After that, the movement of glowworm is governed by the following equation:

(4)

where (t+1) denotes the position of glowworm at t+1 th iteration, s shows the step size and denotes the euclidean distance operator. which takes the luciferin content of the glowworms.

Neighborhood Range Update Phase: Each glowworm is assigned a range which changes with each iteration so that it doesn’t fall for local optimum and can efficiently reach the global optimum. Therefore GSO uses adaptive neighbourhood range to detect multiple peaks of the function. The updation of range equation is shown as:

(5)

where denotes the dynamic range of i th glowworm at t+1 th iteration, denotes the minimum sensory radius, is a constant and is a parameter for controlling the neighbourhood.

Figure 2. Glowworm Swarm Optimization Algorithm

3.0.3. Invasive Weed Optimization (IWO)

This is also a swarm intelligence based algorithm. It is inspired by the colonizing property of the weeds. It was initially presented by Mehrabian and Lucas in 2006. The primary motivation of this algorithm is from the advance and development characteristics of the weed plants that grow in the field. These plants don’t require many resources and exploit the available resources to grow and reproduce. This is the reason for their existence to date, even if humans want to tear them off the field. Weeds have proved very vigorous and adaptable to environmental changes. They have a high survival probability and capable of adaptation.

In the modern agricultural system, many resources are left in the field, creating a budding spot for the weed plant. Weeds take advantage of this and invade the field suing these resources. They colonize, reproduce and disperse new seeds based on their adaptability to nature. The weeds having higher fitness as compared to conventional crops are capable of generating a large number of seeds. These three steps continue until all field reserves are depleted, leaving behind the weeds with high adaptability. This hard competition amongst the plants force them to get adapted with the environment making them more robust to the changes.

The IWO algorithm is generally used to solve several optimization issues using the fashion mentioned above. Compared with some algorithms, it has some special features like colonize, disperse, reproduce and competitive exclusion. Figure 3 shows a flowchart of the IWO. Following are the phases to implement IWO:

  • Initialization:
    Various parameters like population size, number of iterations, constants related to IWO in the search space.

  • Initializing a population:
    All the weeds present initially in the solution space generate an initializing path to the destination from the start. This solution is enhanced in the further iterations.

  • Reproduction:
    Once an initial solution has been formulated, each one is enhanced. Based on the fitness values of each weed, it generates seeds. Compared to less fitting weeds, fitter plants generate more seed .There is a minimum and maximum cap on number of seeds produced per weed plant.

  • Spatial dispersal:

    Once the seeds are created, they are spread to the nearby position around the parent. This ensures that more seed is produced around the parent. Below equation represents the change of variance with respect to the iterations.


    (6)

    where, and iter depict the maximum number of iterations and current iteration. The and

    are the final and initial standard deviations and n is a nonlinear modulation index.


  • Competitive exclusion:
    Once all the process is completed, the weeds are sorted by the fitness value. The total number of weeds exceed the population size, the further population is discarded from the following iterations. The fitter plants produce more seeds to contribute there role in the solution. This instrument utilizes the standard of survival of the fittest strategy.

Figure 3. Invasive Weed Optimization Algorithm

3.0.4. Biogeography-Based Optimization (BBO)

This technique was formulated in 2008 by Dan Simon (Simon, 2008). The above-stated technique is based on the principles of bio-geography. It is the study of the geographical distribution of biological organisms.

3.0.5. Cost Function

Cost function is a function of various factors like length, velocity, height etc. which is used to govern the direction of the robot. The goal of the path planning problem is to create a feasible and safe path, which is achieved by either minimizing or maximizing the cost.The goal of this work is to minimize the cost function.

In the proposed approach, since there is a randomness in the algorithm, there might be a case that the generated path doesn’t reach the target. Therefore the cost function is chosen as follows:

  • is the cost incurred due to travelling from source to destination. The shorter the path, the less is the cost and thus more feasible it is for the robot to take that path.

  • : It is the cost due to sharp turns during the journey. The trajectory has to be smooth since sharp turns leads to more fuel consumption, thereby increasing the energy input of the robot.

  • is the cost incurred due to not reaching the destination due to getting stopped by a dominant obstacle. This is the most important cost and it should be made zero for achieving the best cost.

Thus the total cost function can be shown as follows:

(7)

where , , are constants.

3.0.6. Assumption

Some assumptions had been made in the research to get more efficient and useful results. The assumptions made are as follows:

  • The robot is considered as a point object.

  • The speed of the robot is kept constant during the course of entire simulation. The robot moves faster than compared to targers and obstacles.

  • Upon reaching the target point, a robot can stop immediately irrespective of the momentum of the robot.

4. Results

4.1. Experiment design

4.1.1. Environment

For starting the path planning problem, an environment needs to be set-up/simulated in which the robot can move. There are several properties associated with the environment like obstacles, target, number of robots etc. We have performed simulation using different environment in order to gain more clarity on the performance of the algorithm. There are two types of static obstacles, one that are created in advance and other that is created during run time. These obstacles mimic the real life scenario and can be though of as a building, tree, birds, airplane etc. For simplicity purpose the shape of the obstacles are taken to be cuboidal. The dynamic obstacles are point objects and move randomly in the search space. The robot is assumed to be a point object which can fly in 3D space. There can be multiple goals of different nature like stationary, moving and time varying. The 3D space is considered to be a grid, where each element of the grid is occupied if it contains an obstacle and free otherwise.

Parameter

Value



Map Size
Maximum Path Finding Time
Max percent of static obstacles 8 percent
Total targets

Table 1. Default testing parameters
Figure 4. Environment having no obstacles
Figure 5. Environment having Fixed Static Obstacles(Cuboid)
Figure 6. Environment having Fixed and Runtime Static(small cuboid) Obstacles
Figure 7. Environment having Dynamic Obstacles(black dot)
Figure 8. Environment having Dynamic and Runtime Static Obstacle

Following represents the different components in the map:

  • Green Dot: Robot.

  • Blue Dots: Time varying Goal.

  • Red Dots: Static Goal.

  • Yellow Dots: Dynamic Goal.

  • Black Dots: Dynamic Obstacles.

  • Grey Cuboid: Static Obstacles(FIO).

  • Grey small Cube: Runtime Static Obstacles(FRO).

4.1.2. Experiment description

The simulation has been performed in Python 2.7.9. All the simulations were done on computer with i7-2600 CPU @ 3.40GHz with 4GB RAM. The constants for the simulation are given in Table 1.

4.2. Experiment 1

The first experiment consist of all the algorithms namely GSO, hGSO111here prefix h stands for use of hierarchical levels, IWO, hIWOfootnotemark: , BBO and hBBOfootnotemark: are run in five different environments and cost incurred during path planning is noted. The path followed by the agent is shown in Figures 4-8 and the results can be seen in Tables 2 to 6. Table 2 shows the results for Map 1, which does not consist of any obstacle i.e. all nodes are free nodes. Table 3 shows the results for Map 2, which includes only one type of static obstacles known as FIO(Fixed Individual Obstacles). The position and dimensions of these obstacles are fixed. Table 4 tabulates the results for Map 3, which consists of static obstacles of two types, FIO, and FRO(Fixed Random Obstacles). FRO are generated at runtime and does not have any fixed positions and dimensions. The maximum percentage of nodes covered FRO that can be generated is limited to 8%, as given in Table 1. Table 5 shows the results for Map 4, which consists of both dynamic and static obstacles(only FRO). Dynamic obstacles for the purposes of the simulation are considered point objects, are independent of each other, and their speeds are always less than the speed of the agent. Finally, Table 6 shows the results on Map 5, which contains dynamic obstacles as well as static obstacles of both the kinds(FIO and FRO).

AlgorithmMetric Elapsed time(in sec) Expanded Nodes Cost

GSO
11683 18488 4928032
hGSO 8811 25215 4482761
IWO 32945 77200 11127117
hIWO 79798 122954 10503436
BBO 20574 64268 4176324
hBBO 27401 82045 6296116

Table 2. Simulation results for Map 1
AlgorithmMetric Elapsed time(in sec) Expanded Nodes Cost

GSO
9957 19455 5034259
hGSO 10737 26568 4532053
IWO 65392 97607 11442094
hIWO 65931 120754 8671267
BBO 22611 78290 4335125
hBBO 32329 83000 4290042

Table 3. Simulation results for Map 2
AlgorithmMetric Elapsed time(in sec) Expanded Nodes Cost

GSO
10710 25729 4978856
hGSO 11141 26700 5024262
IWO 60908 102656 11970281
hIWO 85845 128584 10390973
BBO 26511 82502 5256216
hBBO 34931 92818 4766282

Table 4. Simulation results for Map 3
AlgorithmMetric Elapsed time(in sec) Expanded Nodes Cost

GSO
13703 27837 4850499
hGSO 10094 25279 4507687
IWO 65050 95305 8465725
hIWO 85286 130057 11139622
BBO 21972 82746 3952487
hBBO 37490 98404 7572836

Table 5. Simulation results for Map 4
AlgorithmMetric Elapsed time(in sec) Expanded Nodes Cost

GSO
11880 24200 5164557
hGSO 10847 24465 4992658
IWO 78163 111094 12570570
hIWO 84452 116774 10294457
BBO 41883 100409 5364178
hBBO 32184 86921 4299352

Table 6. Simulation results for Map 5

4.3. Experiment 2

The second set of experiments are performed on Map 5, which has dynamically, and obstacles generated randomly at runtime. The results of different algorithms are compared in Table 7-9. The experiments are performed on Map 5 for 30%, 25%, 20% obstacle density for Table 7-9, respectively.

AlgorithmMetric Elapsed time(in sec) Expanded Nodes Cost

GSO
7132 23423 4743890
hGSO 10522 31077 4564806
IWO 45957 130538 13190259
hIWO 43780 123058 11576077
BBO 28994 88954 5289164
hBBO 40917 119277 5487862

Table 7. 30% Obstacle Density
AlgorithmMetric Elapsed time(in sec) Expanded Nodes Cost

GSO
1632 24013 4955365
hGSO 3048 33368 5267612
IWO 15268 82252 6551911
hIWO 24451 120535 10661109
BBO 14793 106620 5255228
hBBO 28616 123523 5183095

Table 8. 25% Obstacle Density
AlgorithmMetric Elapsed time(in sec) Expanded Nodes Cost

GSO
2714 25692 5481396
hGSO 3094 29198 5034207
IWO 17813 110994 10362583
hIWO 23787 122169 8515086
BBO 58434 97418 6092502
hBBO 23921 100271 5963980

Table 9. 20% Obstacle Density

5. Conclusion and future scope

We hereby conclude that we have successfully developed a system that predicts the optimised path based on the environment. The hierarchical version of GSO has proved to give better cost results, in most of the cases, when compared to the normal GSO. hGSO is also better when compared to IWO and BBO and there respective hierarchical versions. The research can be extended to different swarm optimization algorithms communicating with each other and searching the target. Other NIA can also be applied in order to solve the problem of path planning. The multi robot coordination can be enhanced by diving the environment into different regions, with each region having different number of robots which interact with robots in same cluster as well as robots outside the cluster.

References

  • E. U. Acar, H. Choset, Y. Zhang, and M. Schervish (2003) Path planning for robotic demining: robust sensor-based coverage of unstructured environments and probabilistic methods. The International journal of robotics research 22 (7-8), pp. 441–466. Cited by: §1.
  • A. Albert and L. Imsland (2018) Survey: mobile sensor networks for target searching and tracking. Cyber-Physical Systems, pp. 1–42. Cited by: §1.
  • G. Bilbeisi, N. Al-Madi, and F. Awad (2015) PSO-ag: a multi-robot path planning and obstacle avoidance algorithm. In Applied Electrical Engineering and Computing Technologies (AEECT), 2015 IEEE Jordan Conference on, pp. 1–6. Cited by: §2.
  • G. Chaudhary and A. Sinha (2012) Detecting a target location using a mobile robot with range only measurement. In 7th IEEE Conference on Industrial Electronics and Applications (ICIEA), 2012, pp. 28–33. Cited by: §2.
  • Y. Cui, B. Zhang, W. Yang, Z. Wang, Y. Li, X. Yi, and Y. Tang (2017)

    End-to-end visual target tracking in multi-robot systems based on deep convolutional neural network

    .
    In

    Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition

    ,
    pp. 1113–1121. Cited by: §2.
  • M. Dadgar, S. Jafari, and A. Hamzeh (2016) A pso-based multi-robot cooperation method for target searching in unknown environments. Neurocomputing 177, pp. 62–74. Cited by: §2.
  • D. Drake, S. Koziol, and E. Chabot (2018) Mobile robot path planning with a moving goal. IEEE Access 6, pp. 12800–12814. Cited by: §2.
  • S. Hayat and Z. Kausar (2015) Mobile robot path planning for circular shaped obstacles using simulated annealing. In Control, Automation and Robotics (ICCAR), 2015 International Conference on, pp. 69–73. Cited by: §2.
  • U. Jain, W. W. Godfrey, and R. Tiwari (2017)

    A hybridization of gravitational search algorithm and particle swarm optimization for odor source localization

    .
    International Journal of Robotics Applications and Technologies (IJRAT) 5 (1), pp. 20–33. Cited by: §2.
  • A. Janis and A. Bade (2016) Path planning algorithm in complex environment: a survey. Transactions on Science and Technology 3 (1), pp. 31–40. Cited by: §2.
  • E. A. Jensen and M. Gini (2018) Online multi-robot coverage: algorithm comparisons. In Proceedings of the 17th International Conference on Autonomous Agents and MultiAgent Systems, pp. 1974–1976. Cited by: §2.
  • M. Juliá, A. Gil, and O. Reinoso (2012) Searching dynamic agents with a team of mobile robots. Sensors 12 (7), pp. 8815–8831. Cited by: §2.
  • G. Kantor, S. Singh, R. Peterson, D. Rus, A. Das, V. Kumar, G. Pereira, and J. Spletzer (2003) Distributed search and rescue with robot and sensor teams. In Field and Service Robotics, pp. 529–538. Cited by: §1.
  • J. Levinson, J. Askeland, J. Becker, J. Dolson, D. Held, S. Kammel, J. Z. Kolter, D. Langer, O. Pink, V. Pratt, et al. (2011) Towards fully autonomous driving: systems and algorithms. In 2011 IEEE Intelligent Vehicles Symposium (IV), pp. 163–168. Cited by: §1.
  • Z. Liu, X. Mao, and S. Yang (2018) AutoRobot: a multi-agent software framework for autonomous robots. IEICE TRANSACTIONS on Information and Systems 101 (7), pp. 1880–1893. Cited by: §2.
  • A. Marjovi, J. G. Nunes, L. Marques, and A. de Almeida (2009) Multi-robot exploration and fire searching. In Intelligent Robots and Systems, 2009. IROS 2009. IEEE/RSJ International Conference on, pp. 1929–1934. Cited by: §1.
  • E. Masehian and M. Amin-Naseri (2006) A tabu search-based approach for online motion planning. In Industrial Technology, 2006. ICIT 2006. IEEE International Conference on, pp. 2756–2761. Cited by: 2nd item.
  • M. Meghjani, S. Manjanna, and G. Dudek (2016) Multi-target search strategies.. In SSRR, pp. 328–333. Cited by: §2.
  • P. K. Mohanty and D. R. Parhi (2014) A new efficient optimal path planner for mobile robot based on invasive weed optimization algorithm. Frontiers of Mechanical Engineering 9 (4), pp. 317–330. Cited by: §2.
  • A. Pandey, R. K. Sonkar, K. K. Pandey, and D. Parhi (2014) Path planning navigation of mobile robot with obstacles avoidance using fuzzy logic controller. In Intelligent Systems and Control (ISCO), 2014 IEEE 8th International Conference on, pp. 39–41. Cited by: §2.
  • P. Pandey, A. Shukla, and R. Tiwari (2018) Three-dimensional path planning for unmanned aerial vehicles using glowworm swarm optimization algorithm. International Journal of System Assurance Engineering and Management 9 (4), pp. 836–852. Cited by: 1st item, §2.
  • M. Phillips and M. Likhachev (2011) Sipp: safe interval path planning for dynamic environments. In Robotics and Automation (ICRA), 2011 IEEE International Conference on, pp. 5628–5635. Cited by: §2.
  • C. Premebida and U. Nunes (2006) A multi-target tracking and gmm-classifier for intelligent vehicles. In Intelligent Transportation Systems Conference, 2006. ITSC’06. IEEE, pp. 313–318. Cited by: §2, §2.
  • D. Saldana, R. Assunção, and M. F. Campos (2015) A distributed multi-robot approach for the detection and tracking of multiple dynamic anomalies. In Robotics and Automation (ICRA), 2015 IEEE International Conference on, pp. 1262–1267. Cited by: §2.
  • K. Schilling and C. Jungius (1996) Mobile robots for planetary exploration. Control Engineering Practice 4 (4), pp. 513–524. Cited by: §1.
  • A. P. Shukla and R. Tiwari (2017) Discrete problems in nature inspired algorithms. CRC Press. Cited by: §1.
  • D. Simon (2008) Biogeography-based optimization.

    IEEE transactions on evolutionary computation

    12 (6), pp. 702–713.
    Cited by: §3.0.4.
  • G. Wang, H. E. Chu, and S. Mirjalili (2016) Three-dimensional path planning for ucav using an improved bat algorithm. Aerospace Science and Technology 49, pp. 231–238. Cited by: §2.
  • S. Zhang, M. Liu, X. Lei, Y. Huang, and F. Zhang (2018) Multi-target trapping with swarm robots based on pattern formation. Robotics and Autonomous Systems 106, pp. 1–13. Cited by: §2, §2.
  • W. Zhang, X. Gong, G. Han, and Y. Zhao (2017) An improved ant colony algorithm for path planning in one scenic area with many spots. IEEE Access 5, pp. 13260–13269. Cited by: §2.