2D Lidar-Based SLAM and Path Planning for Indoor Rescue ...ruggedsurface,IMUisalsoinaccuratewhichmeansCar-tographermaygetbadresults.So,wechooseHector-SLAM. 4. Path Planning Toachievepathplanningistosolvethreebasicproblems:

Research Article2D Lidar-Based SLAM and Path Planning for Indoor RescueUsing Mobile Robots

Xuexi Zhang 1 Jiajun Lai1 Dongliang Xu1 Huaijun Li 2 and Minyue Fu3

1School of Automation Guangdong University of Technology and Guangdong Key Laboratory of IoT Information TechnologyGuangzhou 510006 China2School of Automobile and Engineering Machinery Guangdong Communication Polytechnic No 789 Tianyuan RoadTianhe District Guangzhou 510630 China3School of Electrical Engineering and Computer Science 0e University of Newcastle University Drive Callaghan2308 NSW Australia

Correspondence should be addressed to Huaijun Li lhjgdcpcn

Received 23 June 2020 Revised 29 July 2020 Accepted 21 October 2020 Published 17 November 2020

Academic Editor Zhiguang Cao

Copyright copy 2020 Xuexi Zhang et al is is an open access article distributed under the Creative Commons Attribution Licensewhich permits unrestricted use distribution and reproduction in any medium provided the original work is properly cited

As the basic system of the rescue robot the SLAM system largely determines whether the rescue robot can complete the rescuemission Although the current 2D Lidar-based SLAM algorithm including its application in indoor rescue environment hasachieved much success the evaluation of SLAM algorithms combined with path planning for indoor rescue has rarely beenstudied is paper studies mapping and path planning for mobile robots in an indoor rescue environment Combined with pathplanning algorithm this paper analyzes the applicability of three SLAM algorithms (GMapping algorithm Hector-SLAM al-gorithm and Cartographer algorithm) in indoor rescue environment Real-time path planning is studied to test the mappingresults To balance path optimality and obstacle avoidance Alowast algorithm is used for global path planning and DWA algorithm isadopted for local path planning Experimental results validate the SLAM and path planning algorithms in simulated emulatedand competition rescue environments respectively Finally the results of this paper may facilitate researchers quickly and clearlyselecting appropriate algorithms to build SLAM systems according to their own demands

1 Introduction

Mobile robots are capable of moving around in their en-vironment and carrying out intelligent activities autono-mously thus having extensive realistic applicationsincluding rescue works A key enabling technology is si-multaneous localization and mapping (SLAM) which allowsthe robot to estimate its own position using onboard sensorsand construct a map of the environment at the same timeWith the SLAM technology real-time path planning can beperformed to fulfill complex manoeuvring tasks in rescueworks

SLAM-enabled mobile robots have achieved muchsuccess in various scenarios Peng et al [1] studied thepositioning problem and implementation of SLAM formobile robots with RGB-D cameras Shou et al [2]

employed a Raspberry Pi module as the core controller andbuilt a mobile robot for map construction and navigation inindoor environment Zhang et al [3] proposed path pre-diction planning based on the artificial potential field toimprove obstacle avoidance Liu et al [4] combined the Q-learning algorithm with the deep learning algorithm for pathplanning which enabled robots to make reasonable walkingpaths under complex environmental conditions Yu et al [5]applied an improved Alowast path planning algorithm to un-manned underwater survey ships enabling quick obstacleavoidance and return to the preset route However thesestudies did not take into account the impact of the rescueenvironment on the SLAM algorithm If these algorithms aredirectly applied to rescue robots it may deteriorate theaccuracy of path planning and even cause incorrect pathplanning results At present there are still rare systems that

HindawiJournal of Advanced TransportationVolume 2020 Article ID 8867937 14 pageshttpsdoiorg10115520208867937

can combine SLAM and path planning for indoor rescueerefore it is necessary to study the impact of the rescueenvironment on the SLAM algorithm and path planningalgorithm and evaluate and select the SLAM algorithmsuitable for the rescue environment

In this paper we evaluated the results of some commonlyused SLAM algorithms in both simulation and real-worldenvironment tested the path planning algorithms (Alowast al-gorithm and DWA algorithm) and conducted a combinedexperiment of mapping and path planning regarding theRoboCup competition ese experiments revealed thedemerits of some algorithms and provided a benchmark forsubsequent algorithm improvement

e rest of the paper is organized as follows Section 2introduces the basic system structure of mobile robotsSection 3 presents the rationale of three commonly usedSLAM algorithms Section 4 briefly describes the Alowast algo-rithm and DWA algorithm for path planning Section 5provides and analyzes the experimental and simulationresults Section 6 gives the conclusion

2 System Structure

e hardware part of the robot studied in this paper ismainly composed of motion control module Lidar modulevision module power module and industrial computermodule e system structure is shown in Figure 1 and thephysical map of the robot system is shown in Figure 2

e main function of the STM32 microcontroller is toacquire and process wheel encoder data and gyroscope dataMap information path planning depth camera and Lidardata are processed by the industrial computere industrialcomputer and STM32 are connected via USB cable to ex-change data and instructions e depth camera is calibratedby using a printed black and white checkerboard eOpenCV function called by the robot operating system(ROS) is used to extract the corner information from cameraimages and then internal and external parameters are ob-tained through calculations [6] e industrial computer isfitted with Intel Core i5 processor 4G memory 128G accessspace and the ubuntu1604 system

3 SLAM Algorithms

For a mobile robot SLAM involves both localization andmapping in an iterative manner by continuously fusingvarious measurements from the onboard sensors [7 8] esensor module in our system includes Lidar and depthcamera to collect environmental information as well asinternal measurements from the IMU A 2D map is to begenerated after processing by a mapping algorithmDepending on the purpose of the map different SLAMalgorithms are available For our purpose we will focus onthe task of path planning in real time After various con-siderations we decide to study in detail three most suitableSLAM algorithms GMapping algorithm Hector-SLAMalgorithm and Cartographer algorithm GMapping algo-rithm is based on particle filter pairing algorithm Hector-SLAM is based on scan matching algorithm Cartographer is

a scan matching algorithm with loop detection and RGB-Dalgorithm is an algorithm for mapping using depth imagesese several algorithms are representative and widely usedalgorithms

31 GMapping Algorithm e GMapping algorithm is alaser-based SLAM algorithm for grid mapping [9 10]is isprobably the most used SLAM algorithm currently thestandard algorithm on the PR2 (a very popular mobilemanipulation platform) with implementation available onopenslamorg e algorithm was initially proposed in [10]and the main idea is to use RaondashBlackwellized particle filters(RBPFs) to predict the state transition function e algo-rithm is also known as the RBPF SLAM algorithm namedafter the use of RaondashBlackwellized particle filters In [11]two major improvements were made by optimizing theproposal distributions and introducing adaptive resamplingmaking the algorithm much more suitable for practicalapplications It is then dubbed GMapping (G for grid) due tothe use of grid maps

311 RBPF Onboard measurements include sensor datafrom Lidar or camera for images and odometer data fromthe IMU A large number of particles are used for statetransition function predictions with each particle repre-senting a possible position of the robot

e sensor data are denoted by (z1t z1 z2 zt) andthe odometer data by (u1t u1 u2 utminus1) for the timeperiod from 1 to t ey are used to estimate the jointposterior probability p(x1t m|z1t u1tminus1) of the robot pose(x1t x1 x2 xt) and the grip map of the environmentrepresented by m Using the Bayesrsquo rule the posteriorprobability can be decomposed into

p x1t m|z1t u1tminus1( 1113857

p x1t|z1t u1tminus1( 1113857p m|x1t z1t( 1113857(1)

where p(x1t|z1t u1tminus1) is the positioning problem whereasp(m|x1t z1t) is the mapping problem e so-called im-portance sampling is used in the RBPF e procedure is asfollows

(i) Sampling according to the given (previous) proposaldistribution particles (x

(i)tminus1) from the previous

generation are sampled ey are then improved byincorporating the most recent observations ennew particles (x

(i)t ) and proposal distributions are

generated(ii) Weights the weight w

(i)t of each current particle x

(i)t

is calculated using

w(t)t

p x(i)1t|z1t u1t1113872 1113873

π x(i)1t|z1t u1tminus11113872 1113873

(2)

where π(middot) is the proposal distribution whichusually is a probabilistic odometry motion model

2 Journal of Advanced Transportation

(iii) Resampling depending on the weights particleswith smaller weights are discarded and replaced byresampled particles but the total number of par-ticles in the resampled particle set is unchanged

(iv) Map updating the map update is implemented bythe pose represented by each particle in combina-tion with the current observation To reduce thecomputational complexity a recursive formula forweight update is used

w(i)t w

(i)tminus1η

p zt|x(i)1t z1tminus11113872 1113873p x

(i)t |x

(i)1tminus1 u1tminus11113872 1113873

π x(i)1t|z1t u1tminus11113872 1113873

(3)

where η is a normalisation factor

312 Proposal Distribution A large number of particles willcause a large amount of calculation andmemory consumptionIn order to reduce the number of particles a proposal dis-tribution is used Our target distribution is the best distributionof the robot state according to the data of all sensors carried bythe robot Except for the odometermodel the laser observationdata is the position information of 360-degree points which isdifficult to performGaussianmodellingus there is no directway to sample the target distribution and the proposal dis-tribution is used instead of the target distribution to extract therobot pose information at the next time instant e proposaldistribution considers not only the motion (odometer) in-formation but also the most recent observation (laser) infor-mationis canmake the proposal distribution more accurateand closer to the target distribution

Power supplysystem

Motor controlmodule

STM32 robotcontrol module IPC Lidar

Encoder Gyroscope RGB-D

WiFi

PC

Figure 1 System structure of our mobile robot ldquoSTM32 robot control modulerdquo gets information of motors from ldquoencoderrdquo and motionfrom ldquogyroscoperdquo to control motors and transfer the motion and encoder information to ldquoIPCrdquo which is a microcomputeren ldquoIPCrdquo getsinformation from Lidar ldquoSTM32 robot control modulerdquo and RGB-D camera

(a) (b)

Figure 2 Key modules of the robot (1) ldquoRGB-Drdquo used to get RGB image and depth image (2) ldquolidarrdquo used to get 2D lidar point cloud (3)ldquoSTM32 controllerrdquo used to control the movement of the car (4) ldquogyroscoperdquo used to get the attitude information (5) ldquoWI-FIrdquo used to contactwith the host computer (6) ldquopower supply systemrdquo and (7) ldquomotor and encoderrdquo used to get the velocity feedback of the car movement

Journal of Advanced Transportation 3

e sensor observation information is added whencalculating the proposal distribution and the samplingprocess is concentrated in the peak region of the like-lihood function to get the optimal proposal distribution

p xt|x(i)t m

(i)tminus1 zt utminus11113872 1113873

p zt|xt m(i)

1113872 1113873p xt|x(i)tminus1 utminus11113872 1113873

p zt|x(i)tminus1 m

(i) utminus11113872 1113873

(4)

en the weights are updated according to the aboveweight recursion formula

w(i)t w

(i)tminus1η

p zt|xt m(i)


p x(i)t |x

(i)tminus1 m

(i) zt utminus11113872 1113873

propw(i)tminus1η

p zt|xt m(i)


p zt|x(i)tminus1 m

(i) utminus11113872 1113873

w(i)tminus1p zt|x

(i)tminus1 m

(i) utminus11113872 1113873

(5)

e Gaussian distribution is used to approximate theapproximated peak region of the observation andthe optimal proposal distribution is obtained eGaussian distribution parameters ie the means μ(i)

t

and covariances 1113936(i)t are determined using K sampling

points

μ(i)t

1η(i)

1113944

K

j1xjp zt|xj m

(i)tminus11113872 1113873p xj|x

(i)tminus1 utminus11113872 1113873

1113944(i)

t

1η(i)

1113944

K

j1p zt|xj m

(i)tminus11113872 1113873p xj|x


middot xj minus μ(i)t1113872 1113873 xj minus μ(i)

t1113872 1113873T

(6)

where the normalising factor η(i) is given by

η(i) 1113944

K

j1xjp zt|xj m

(i)tminus11113872 1113873p xj|x

(i)tminus1 utminus11113872 1113873 (7)

313 Adaptive Resampling Resampling may cause goodparticles to be removed from the filter making the par-ticles scarce erefore it is necessary to judge the qualityof the particles by the effective sampling scale standardand judging the time of resampling e evaluation for-mula is as follows

Neff 1

1113936Ni1 1113957w

(i)1113872 1113873

2 (8)

where N is the number of particles and 1113957w(i) is the weight ofthe ith particlee worse the proposal distribution estimatethe smaller the Neff is When Neff lt (12)N GMappingperforms resampling

32 Hector-SLAM Algorithm e Hector-SLAM algorithm[12] differs from other grid-based mapping algorithms as itdoes not require odometer information but it needs laserdata and a priori map Hector-SLAM is based on theGaussndashNewton iteration formula that optimally estimatesthe pose of the robot as represented by the rigid bodytransformation ξ [px pyψ]T from the robot to the priormap e optimal estimation is done by optimally matchingthe laser data and the map in the sense that the optimal ξlowast

below is solved

ξlowast argminξ 1113944

N

i11 minus M Si(ξ)( 11138571113858 1113859

2 (9)

Here M(Si(ξ)) is the value of the map at Si(ξ) and Si(ξ)

is the world coordinate of scan end points si (six xiy)Twhich obeys the following function

Si(ξ) cosψ minussinψ

sinψ cosψ1113890 1113891

six

siy

⎡⎣ ⎤⎦ +px

py

⎡⎣ ⎤⎦ (10)

When an initial estimate of the pose ξ is given anupdated estimate ξ + Δξ is computed by approximatingM(Si(ξ + Δξ)) using first-order Taylor expansion and theresult is as follows

Δξ Hminus 1

1113944

N

i1nablaM Si(ξ)( 1113857

zSi(ξ)

zξ1113890 1113891

T

1 minus M Si(ξ)( 11138571113858 1113859 (11)

where H is the Hessian matrix or some approximation of itgiven by

H nablaM Si(ξ)( 1113857zSi(ξ)

zξ1113890 1113891

T

nablaM Si(ξ)( 1113857zSi(ξ)

zξ1113890 1113891 (12)

33 Cartographer Algorithm When the amount of data toprocess becomes too large particle-based algorithms are notapplicable due to their higher computing requirements onthe processor In this case graph optimisation algorithmsare more suitable

Googlersquos solution to SLAM called Cartographer is agraph optimisation algorithm e Google open sourcecode1 consists of two parts Cartographer and Cartogra-pher_ROS e function of Cartographer is to process thedata from Lidar IMU and odometers to build a mapCartographer_ROS then acquires the sensor data throughthe ROS communicationmechanism and converts them intothe Cartographer format for processing by Cartographerwhile the Cartographer processing result is released fordisplay or storage Impressive real-time results for solvingSLAM in 2D have been described in [13] by the authors ofthe software

34 Considering the Rescue Environment In rescue envi-ronment there are stairs and rugged surface which make theodometer inaccurate It means we could not chooseGMapping because it is very rely on odometer Due to the


rugged surface IMU is also inaccurate which means Car-tographer may get bad results So we choose Hector-SLAM

4 Path Planning

To achieve path planning is to solve three basic problems

(1) e robot reaches the desired position(2) e obstacle avoidance and completion of the

strategic task are achieved in the moving process(3) e optimal path is realized

However in the actual environment due to the accuracyof the robot sensor and the variability of the environmentthe environmental information and location information ofthe map construction will be deviated [14] Global planningin a static environment can meet the problem requirementsbut to handle the deviation caused by the dynamic envi-ronment local path planning needs to be introducedat isto say the local path planning pays more attention to ob-stacle avoidance and the global path planning pays moreattention to the shortest path erefore the combination oflocal planning and global planning algorithms can suc-cessfully achieve accurate navigation of the robot e globalplanning algorithm studied in this paper is a node-based Alowast

algorithm [15 16] and the local planning algorithm is adynamic window algorithm (DWA) [17] Global pathplanning produces a high-level plan for the robot to follow toreach the goal location Local path planning is responsiblefor generating velocity commands for the mobile unit tosafely move the robot toward a goal ese properties areimbedded in the plan produced by the planners using thecost function which takes into account both distance toobstacles and distance to the path

41GlobalPathPlanning Based on the global path planningof the grid method the Alowast algorithm is used to study thepath planninge Alowast algorithm follows the cost function tomake the robot to directionally search for the path towardthe end point e core valuation function of the Alowast al-gorithm is

f(n) g(n) + h(n) (13)

where the node n is abstractly understood as the next targetpoint f(n) represents the total valuation function of thecurrent node n g(n) represents the actual cost of the startingpoint to the current point and h(n) represents the estimatedcost of the current node to the end point e value of h(n)

determines the performance of the algorithm Typicallyh(n) uses the Euclidean distance or Manhattan distancebetween the two points in space In the Alowast algorithm the

Manhattan distance is used e Manhattan distance be-tween two points (x1 y1) and (x2 y2) is as follows

DManhattan x1 minus x21113868111386811138681113868

1113868111386811138681113868 + y1 minus y21113868111386811138681113868

1113868111386811138681113868 (14)

42 Local Path Planning is paper mainly studies thecorresponding action strategy and operation of robot forpath navigation in indoor environment For this reason theDWA algorithm is selected as the main algorithm for localpath planning e DWA algorithm requires the robot toperform numerical simulation calculations on the path ofthe robot within a certain speed windowus it is necessaryto obtain the model state expression of the robot e two-wheeled robot based on differential drive has no velocity inthe y-axis direction Since the robot is at the millisecondlevel in each sampling period of the program execution themotion trajectory of the robot in the two adjacent samplingperiods can be approximated as a straight line In a period oftime Δ the robot travels a small distance at speed v and it isat an angle θt to the x-axis then the movement incrementsΔx and Δy of the robot on the x-axis and the y-axis can beobtained respectively

Δx x + vΔt cos θt( 1113857 (15)

Δy y + vΔt sin θt( 1113857 (16)

e robotrsquos movement trajectory is then given by

xt+1 xt + vΔt cos θt( 1113857 (17)

yt+1 yt + vΔt sin θt( 1113857 (18)

θt+1 θt + ωΔt (19)

where ω is the angular velocity of the robotDuring the speed sampling of the robot multiple sets of

trajectory velocity values are collected To make the robotsafely perform path planning some necessary speed limitsare also needed e speed velocity value and angular ve-locity value of the robot change within a certain range andthe range needs to be empirically calculated according to thephysical characteristics of the robot and the operating en-vironment e range formula is as follows

Vm (vω)|v isin vmin vmax1113858 1113859ω isin ωminωmax1113858 11138591113864 1113865 (20)

e robot has different torque performance parametersdue to different motor models When the current speed ofthe robot vc and the angular velocityωc are known the actualspeed range for the next sampling time can be computed as

Va (vω)|v isin vc minus _vdΔt vc + _vaΔt1113858 1113859ω isin ωc minus _ωdΔtωc + _ωaΔt1113858 11138591113864 1113865 (21)


where _va and _ωa are the maximum accelerations and _vd and_ωd are the maximum decelerations

When the robot is safely avoiding obstacles in naviga-tion the speed (vω) during the whole locally plannedtrajectory must be within the range of speeds given by

Vd (vω)|

2dis(vω) _vd

1113969

ge vc

2dis(vω))ωd

1113969

geωc1113882 1113883

(22)

where dis(vω) is the minimum distance from the currentposition to the point where the arc trajectory of v and w

intersects the nearest obstaclePerforming the DWA algorithm for speed selection

needs to satisfy equations (19)ndash(21) simultaneously On thebasis of the trajectory that satisfies these speed requirementsan evaluation function is used to measure a selected tra-jectory and to aim the selection of the optimal trajectoryeevaluation function is as follows

G(vω) σ(αhead(vω)) + βdis(vω) + cvel(vω) (23)

where head(vω) represents the angle difference between theestimated end of the route and the target dis(vω) is theminimum distance from the obstacle to the planned tra-jectory as explained above vel(vω) indicates the momentspeed evaluation σ(middot) is a smoothing function and α β cgt 0are evaluation coefficients

5 Experimental Results

51 Simulation Experiments

511 SLAM Simulation In order to test the aforementionedalgorithms we carry out simulation experiments using theGazebo platform to build the simulation environmentshown in Figure 3e virtual environment has real physicalproperties and the simulation results have strong referenceto the actual environment

e following simulation experiment was performedaccording to the virtual environment We first use Lidar as asensor to simulate the GMapping Hector-SLAM andCartographer algorithms We then use depth camera (RGB-D) as a sensor to simulate the GMapping algorithm Fourmapping algorithms under the physical simulation platformof the ROS robot system are tested and the simulationresults are shown in Figures 4ndash7 e purpose of presentingthese diagrams is to give an impression of the algorithmsdescribed above which are parsed using text and formulas

Figure 4 shows the robot simulation process usingGMapping Depth information of the Lidar is required erobot is located in the lower left corner of simulation en-vironment the data collected by the Lidar is marked in redand the established environment map is in light gray Fig-ure 5 shows the robot simulation process using Hector-SLAM Figure 6 shows the robot simulation process usingCartographer e area scanned by the Lidar changes fromlight gray to white until the whole map is completed

Figure 7 shows the final grid map constructed using theRGB-D camera data e depth data are first transformedusing depthimage_to_laserscan before being applied to

GMappinge simulation results show that the constructedmap is not ideal is is because the RGB-D camera is af-fected by its own structure causing a limited range ofviewing angle [18] In addition the depth camera requiresrich scene features to work e simulated environment hassmooth wall surfaces with very few scene features makingthe depth camera unable to perform effective featurematching We see that map construction cannot be suc-cessfully completed

512 Path Planning Simulation After obtaining the envi-ronment map the map was loaded under the ROS frame-work for path planning purposes We use the rviz packageunder the ROS framework for path planning and navigationsimulation e constructed map is shown in the left part ofFigure 8 Here we also see the black dot indicating anobstacle and the cyan portion indicates the safe distancebetween the robot and the obstacle e green arrow pointsto the target point and direction of the robot e right partof Figure 8 shows the path planning and navigation resultsas indicated by the green line e starting position for therobot is slightly below the obstacleWe see that the simulatedpath not only successfully avoids the obstacle but also is anearly straight path

52 Algorithm Verification in Lab Environment

521 Emulated Rescue Experiment In this experiment a125m times 125m square wooden structure was used tosplicing the actual environment in an open laboratory asshown in Figure 9 is was built with reference to theRoboCup rescue venue to simulate an enclosed indoorrescue environment after a disaster e aforementionedSLAM algorithms are applied and the results are shown inFigure 10 Comparing with the simulation results the mapsconstructed in the actual environment using Lidar data(GMapping Hector-SLAM and Cartographer) are all sat-isfactory In the RGB-D experiment [19 20] the environ-mental features are not sufficient for depth measurementsresulting in an overlapped map erefore the mappingalgorithm based on the depth camera needs furtheroptimisation

522 Lab Office Experiment Next we test the SLAM al-gorithms in a real lab office as shown in Figure 11 e mapin Figure 12 is constructed using the Hector-SLAM algo-rithm We see that not only desks are clearly identified butalso the chair legs are clearly shown

Path planning is carried out using the Alowast algorithm forglobal path planning and DWA algorithm for local pathplanning and the results are shown in Figure 13 In thepicture on the left the navigation target point and directionof the robot are set by the green arrowe small green patchat the bottom shows many arrows representing the particles(their positions and directions) of the starting pose of therobote right picture shows that after the execution of thepath planning and navigation the robot moves to the target


5m

5m

(a) (b)

Figure 3 Simulation environment diagram and Gazebo display

1 2 3 4

Figure 4 Mapping result of gmapping (simulation)

1 2 3 4

Figure 5 Mapping result of Hector-SLAM (simulation)


point as indicated by the (smaller) green patch representingthe particles of the final pose We see that the experimentalresults verify the effectiveness of the path planningalgorithms

523 RoboCup Competition Test e mapping and pathplanning algorithms above are used in our entry of the 2019RoboCup competition for indoor rescue [21] e compe-tition venue is shown in Figure 14 e competition re-

1 2 3 4

Figure 6 Mapping result of Cartographer (simulation)

Figure 7 Mapping result of RBG-D (simulation)


Figure 9 Emulated rescue environment

(a) (b)

Figure 8 Simulation result of path planninge green arrow on the left picture means the orientation of the robot when it reaches the end

(a) (b)

Figure 10 Continued


quirements are as follows robot constructs the map of thecompetition venue by self-exploration and completes therecognition of the doll and the marking of the QR code erobot operates by either remote control or through an au-tonomous exploration algorithm In addition the robotmust avoid obstacles e quality of the mapping result is

assessed by the number of closed grids identified in theconstructed grid map Because GMapping and Cartographerneed IMU to assist positioning and in uneven terrain in thecompetition IMU data will have very large errors whichleads to very large errors in these two SLAM algorithmseRGB-D algorithm also does not perform well in flat terrain

(c) (d)

Figure 10 Constructed map (a) RGB-D (b) Hector-SLAM (c) GMapping (d) Cartographer

Figure 11 Lab office environment

Figure 12 Hector-SLAM map for lab office


so we use the Hector-SLAM algorithm for mapping And weuse the Hector_navigation open source software package[22] for robot self-exploration As shown in Figure 15 due tothe complex environment of the competition venue therobot is unable to pass some obstacles such as the 15-degreeslope and stairs so certain parts of the venue cannot bescanned by the Lidar and the constructed map is incom-plete Due to the limitation of the robot hardware it is notpossible to finish all the mapping But all the places reachedby the robot have been well-mapped Finally we scored 14points (out of 27 points) in the self-exploration session

524 China Robot Competition Compared with theRoboCup competition venue the 2019 China RobotCompetition venue has a larger site area and a larger slopewhich means that the terrain is more complicated as shownin Figure 16 e competition requires rescue robots toindependently explore the map of the field and identify the

two-dimensional code on the box in the simulated post-disaster environment In this experiment we continue to usethe Hector_navigation open source software package forrobotic exploration e route of the robotrsquos autonomousnavigation is shown in Figure 17We used two servos to keepthe Lidar level which is used to automatically adjust theLidar to a horizontal position However due to the fact thatthe competition field is not flat which causes the robot tohave large fluctuations during the movement the Lidar isunable to adjust the pose in time As shown in Figure 18 theyellow arrow indicates the starting point of the rescue robotthe purple line marks the movement trajectory of the rescuerobot the dark blue line represents the map constructed bythe SLAM algorithm on the competition venue and the darkblue circle with numbers represents the location of the QRcode However the robot is navigated outside the wall As aresult there is a positioning error e map cannot bequickly updated for corrections and the navigation algo-rithm may incorrectly navigate the robot into an obstacle

(a) (b)

Figure 14 RoboCup venue

(a) (b)

Figure 13 Path planning for lab office


(a) (b)

Figure 16 China robot competition venue

(a) (b)

Figure 15 Mapping result of Hector-SLAM (RoboCup)

(a) (b)

Figure 17 Continued


erefore the Hector-SLAM algorithm needs to be im-proved in the update speed and the navigation algorithmshould adopt a more cautious strategy according to theapplication of the rescue scenario As shown in Figure 18 anunnecessary closed point appears on the periphery of theconstructed map In addition the number in the figure is theposition information of the recognised QR code

In the above simulation experiments and field experi-ments for GMapping Hector-SLAM Cartographer andRGB-D mapping algorithms GMapping Hector-SLAMand Cartographer perform better in a flat indoor environ-ment e RGB-D mapping algorithm has poor mappingeffect due to poor lighting conditions and lack of envi-ronmental features erefore in a relatively flat indoorrescue environment it is more appropriate to choose thefirst three algorithms as the basis of the SLAM system In theRoboCup competition and the China Robot Competitionthe uneven rescue environment seriously interfered with theIMU data so the SLAM algorithm (GMapping and Car-tographer) incorporating IMU data could not perform

SLAM tasks normally is causes the path planning algo-rithm that relies on map information to not work correctlye Hector-SLAM algorithm which does not rely on IMUdata can perform tasks in the competition terrain relativelycorrectly but there are still problems with inability to filterwrong Lidar information and poor stability

6 Conclusions

In this paper the problem of indoor rescue using mobilerobots was studied Comparisons were done on theGMapping Hector-SLAM and Cartographer algorithms forSLAM e path planning was done by combining the Alowast

algorithm for global path planning and the DWA algorithmfor local path planning Simulation emulation as well as realenvironment experiments were conducted to compare andvalidate the results on map construction and path planningIn the future further optimisation needs to be carried out inthe mapping algorithms to make them more suitable to thereal rescue environment

Figure 18 Mapping result of Hector-SLAM (China Robot Competition) Because the level keeping platform of the radar is not sensitiveenough and it cannot be adjusted and keep the platform in time when the robot crosses obstacles the radar may scan the walls outside thefield Hector cannot judge and filter so it is recorded on the map

(c) (d)

Figure 17 Mapping result and autonomous navigation route (China robot competition) (a) 1 (b) 2 (c) 3 (d) 4


Data Availability

Our experiment data can be found in httpsgithubcom9393dlRescue_Robot

Conflicts of Interest

e authors declare that they have no conflicts of interest

Acknowledgments

is work was supported by the National Natural ScienceFoundation of China (Grant nos 61633014 61803101 andU1701264)

References

[1] W Peng F Yuan and Z Zhou ldquoPositioning analysis andimplementation of mobile robot based on RGB-D camerardquoIntelligent Computer and Applications vol 9 no 3pp 168ndash170 2019

[2] J Shou Z Zhang Y Su and Z Zhong ldquoDesign andimplementation of indoor positioning and navigation systemof mobile robot based on ROS and lidarrdquo Machinery andElectronics vol 36 no 11 pp 76ndash80 2018

[3] H Zhang X Zhang and Q Wang ldquoPath planning based onpath prediction artificial potential field method for automaticfollowing trolleyrdquo Computer Measurement and Controlvol 27 no 1 pp 237ndash240 2019

[4] Z Liu S Jiang W Yuan and C Shi ldquoRobot path planningbased on deep Q-learningrdquo Measurement and ControlTechnology vol 38 no 7 pp 24ndash28 2019

[5] B Yu X Chu C Liu H Zhang and Q Mao ldquoPath planningmethod for unmanned waterway survey ships based on im-proved Alowast algorithmrdquo Geomatics and Information Science ofWuhan University vol 44 no 8 pp 1258ndash1264 2019

[6] S Sugiyama H Okuda S Inagaki and T Suzuki ldquoSLAMusing ROS and operational assist for electric wheelchairsrdquo inProceedings of the JSME Annual Conference on Robotics andMechatronics (Robomec) November 2017

[7] B Sun ldquoMobile robot SLAM technologyrdquo Electronic Tech-nology and Software Engineering vol 1 no 2 p 95 2018

[8] Z Lin and S Zheng ldquoResearch on image matching andcamera pose resolution in mobile robot vision SLAM pro-cessrdquo Machine Design and Manufacturing Engineeringvol 46 no 11 pp 13ndash18 2017

[9] G Grisetti C Stachniss and W Burgard ldquoImproved tech-niques for grid mapping with Rao-Blackwellized particlefiltersrdquo IEEE Transactions on Robotics vol 23 no 1pp 34ndash46 2007

[10] A Doucet J F G de Freitas K Murphy and S Russel ldquoRao-Blackwellized particle filtering for dynamic bayesian net-worksrdquo in Proceedings of the Conference on Uncertainty inArtificial Intelligence (UAI) pp 176ndash183 Stanford CA USAJune 2000

[11] N Koenig and A Howard ldquoDesign and use paradigms forGazebo an open-source multi-robot simulatorrdquo in Pro-ceedings of the 2004 IEEERSJ International Conference onIntelligent Robots and Systems pp 2149ndash2154 Sendai JapanSeptember 2004

[12] S Kohlbrecher O Von Stryk J Meyer and U Klingauf ldquoAflexible and scalable slam system with full 3D motion esti-mationrdquo in Proceedings of the 2011 IEEE International

Symposium on Safety Security and Rescue Robotics KyotoJapan November 2011

[13] W Hess D Kohler H Rapp and D Andor ldquoReal-time loopclosure in 2D Lidar SLAMrdquo in Proceedings of the IEEE In-ternational Conference on Robotics and Automation (ICRA2016) Stockholm Sweden May 2016

[14] X Wang L He and T Zhao ldquoMobile robot for SLAM re-search based on lidar and binocular vision fusionrdquo ChineseJournal of Sensors and Actuators vol 31 no 3 pp 394ndash3992018

[15] P Hart N Nilsson B Raphael et al ldquoA formal basis for theheuristic determination of minimum cost pathsrdquo IEEETransactions on Systems Science and Cybernetics vol 4 no 2pp 100ndash107 1968

[16] Z Cao S Jiang J Zhang and H Guo ldquoA unified frameworkfor vehicle rerouting and traffic light control to reduce trafficcongestionrdquo IEEE Transactions on Intelligent TransportationSystems vol 18 no 7 pp 1958ndash1973 2017

[17] M Seder and I Petrovic ldquoDynamic window based approachto mobile robot motion control in the presence of movingobstaclesrdquo in Proceedings of the 2007 IEEE InternationalConference on Robotics and Automation pp 1986ndash1991Roma Italy April 2007

[18] Z Cao H Guo J Zhang D Niyato and U FastenrathldquoFinding the shortest path in stochastic vehicle routing acardinality minimization approachrdquo IEEE Transactions onIntelligent Transportation Systems vol 17 no 6 pp 1688ndash1702 2016

[19] Y Li S Wang and N Yu ldquoRGB-D-based SLAM and pathplanning for mobile robotsrdquo CAAI Transactions on IntelligentSystems vol 13 no 3 pp 445ndash451 2018

[20] J Cai K Chen and Y Zhang ldquoImproved V-SLAM for mobilerobots based on Kinectrdquo CAAI Transactions on IntelligentSystems vol 13 no 5 pp 734ndash740 2018 httpknscnkinetkcmsdetail231538tp201804230957004html

[21] H Kitano M Asada Y Kuniyoshi I Noda and E OsawaldquoRoboCup the robot world cup initiativerdquo in Proceedings ofthe International Conference on Autonomous Agents MarinaDel Rey CA USA February 1997

[22] S Kohlbrecher J Meyer T Graber K Petersen U Klingaufand O Von Stryk ldquoHector open source modules for au-tonomous mapping and navigation with rescue robotsrdquo inProceedings of the Robot Soccer World Cup pp 624ndash631 JoaoPessoa Brazil January 2014


can combine SLAM and path planning for indoor rescueerefore it is necessary to study the impact of the rescueenvironment on the SLAM algorithm and path planningalgorithm and evaluate and select the SLAM algorithmsuitable for the rescue environment

In this paper we evaluated the results of some commonlyused SLAM algorithms in both simulation and real-worldenvironment tested the path planning algorithms (Alowast al-gorithm and DWA algorithm) and conducted a combinedexperiment of mapping and path planning regarding theRoboCup competition ese experiments revealed thedemerits of some algorithms and provided a benchmark forsubsequent algorithm improvement

e rest of the paper is organized as follows Section 2introduces the basic system structure of mobile robotsSection 3 presents the rationale of three commonly usedSLAM algorithms Section 4 briefly describes the Alowast algo-rithm and DWA algorithm for path planning Section 5provides and analyzes the experimental and simulationresults Section 6 gives the conclusion

2 System Structure

e hardware part of the robot studied in this paper ismainly composed of motion control module Lidar modulevision module power module and industrial computermodule e system structure is shown in Figure 1 and thephysical map of the robot system is shown in Figure 2

e main function of the STM32 microcontroller is toacquire and process wheel encoder data and gyroscope dataMap information path planning depth camera and Lidardata are processed by the industrial computere industrialcomputer and STM32 are connected via USB cable to ex-change data and instructions e depth camera is calibratedby using a printed black and white checkerboard eOpenCV function called by the robot operating system(ROS) is used to extract the corner information from cameraimages and then internal and external parameters are ob-tained through calculations [6] e industrial computer isfitted with Intel Core i5 processor 4G memory 128G accessspace and the ubuntu1604 system

3 SLAM Algorithms

For a mobile robot SLAM involves both localization andmapping in an iterative manner by continuously fusingvarious measurements from the onboard sensors [7 8] esensor module in our system includes Lidar and depthcamera to collect environmental information as well asinternal measurements from the IMU A 2D map is to begenerated after processing by a mapping algorithmDepending on the purpose of the map different SLAMalgorithms are available For our purpose we will focus onthe task of path planning in real time After various con-siderations we decide to study in detail three most suitableSLAM algorithms GMapping algorithm Hector-SLAMalgorithm and Cartographer algorithm GMapping algo-rithm is based on particle filter pairing algorithm Hector-SLAM is based on scan matching algorithm Cartographer is

a scan matching algorithm with loop detection and RGB-Dalgorithm is an algorithm for mapping using depth imagesese several algorithms are representative and widely usedalgorithms

31 GMapping Algorithm e GMapping algorithm is alaser-based SLAM algorithm for grid mapping [9 10]is isprobably the most used SLAM algorithm currently thestandard algorithm on the PR2 (a very popular mobilemanipulation platform) with implementation available onopenslamorg e algorithm was initially proposed in [10]and the main idea is to use RaondashBlackwellized particle filters(RBPFs) to predict the state transition function e algo-rithm is also known as the RBPF SLAM algorithm namedafter the use of RaondashBlackwellized particle filters In [11]two major improvements were made by optimizing theproposal distributions and introducing adaptive resamplingmaking the algorithm much more suitable for practicalapplications It is then dubbed GMapping (G for grid) due tothe use of grid maps

311 RBPF Onboard measurements include sensor datafrom Lidar or camera for images and odometer data fromthe IMU A large number of particles are used for statetransition function predictions with each particle repre-senting a possible position of the robot

e sensor data are denoted by (z1t z1 z2 zt) andthe odometer data by (u1t u1 u2 utminus1) for the timeperiod from 1 to t ey are used to estimate the jointposterior probability p(x1t m|z1t u1tminus1) of the robot pose(x1t x1 x2 xt) and the grip map of the environmentrepresented by m Using the Bayesrsquo rule the posteriorprobability can be decomposed into

p x1t m|z1t u1tminus1( 1113857

p x1t|z1t u1tminus1( 1113857p m|x1t z1t( 1113857(1)

where p(x1t|z1t u1tminus1) is the positioning problem whereasp(m|x1t z1t) is the mapping problem e so-called im-portance sampling is used in the RBPF e procedure is asfollows

(i) Sampling according to the given (previous) proposaldistribution particles (x

(i)tminus1) from the previous

generation are sampled ey are then improved byincorporating the most recent observations ennew particles (x

(i)t ) and proposal distributions are

generated(ii) Weights the weight w

(i)t of each current particle x

(i)t

is calculated using

w(t)t

p x(i)1t|z1t u1t1113872 1113873

π x(i)1t|z1t u1tminus11113872 1113873

(2)

where π(middot) is the proposal distribution whichusually is a probabilistic odometry motion model




w(i)t w

(i)tminus1η


(i)t |x

(i)1tminus1 u1tminus11113872 1113873

π x(i)1t|z1t u1tminus11113872 1113873

(3)



Power supplysystem

Motor controlmodule



WiFi

PC


(a) (b)




p xt|x(i)t m


p zt|xt m(i)


p zt|x(i)tminus1 m

(i) utminus11113872 1113873

(4)


w(i)t w

(i)tminus1η

p zt|xt m(i)


p x(i)t |x

(i)tminus1 m

(i) zt utminus11113872 1113873

propw(i)tminus1η

p zt|xt m(i)


p zt|x(i)tminus1 m

(i) utminus11113872 1113873

w(i)tminus1p zt|x

(i)tminus1 m

(i) utminus11113872 1113873

(5)


t


points

μ(i)t

1η(i)

1113944

K

j1xjp zt|xj m

(i)tminus11113872 1113873p xj|x


1113944(i)

t

1η(i)

1113944

K

j1p zt|xj m

(i)tminus11113872 1113873p xj|x



t1113872 1113873T

(6)


η(i) 1113944

K

j1xjp zt|xj m

(i)tminus11113872 1113873p xj|x

(i)tminus1 utminus11113872 1113873 (7)


Neff 1

1113936Ni1 1113957w

(i)1113872 1113873

2 (8)



below is solved


N

i11 minus M Si(ξ)( 11138571113858 1113859

2 (9)




sinψ cosψ1113890 1113891

six

siy

⎡⎣ ⎤⎦ +px

py

⎡⎣ ⎤⎦ (10)


Δξ Hminus 1

1113944

N


zSi(ξ)

zξ1113890 1113891

T

1 minus M Si(ξ)( 11138571113858 1113859 (11)



zξ1113890 1113891

T


zξ1113890 1113891 (12)






4 Path Planning







f(n) g(n) + h(n) (13)





1113868111386811138681113868 + y1 minus y21113868111386811138681113868

1113868111386811138681113868 (14)


Δx x + vΔt cos θt( 1113857 (15)

Δy y + vΔt sin θt( 1113857 (16)


xt+1 xt + vΔt cos θt( 1113857 (17)

yt+1 yt + vΔt sin θt( 1113857 (18)

θt+1 θt + ωΔt (19)









Vd (vω)|

2dis(vω) _vd

1113969

ge vc

2dis(vω))ωd

1113969

geωc1113882 1113883

(22)



















5m

5m

(a) (b)


1 2 3 4


1 2 3 4





1 2 3 4





(a) (b)


(a) (b)

Figure 10 Continued




(c) (d)








(a) (b)


(a) (b)



(a) (b)


(a) (b)


(a) (b)

Figure 17 Continued





6 Conclusions




(c) (d)



Data Availability




Acknowledgments


References



























w(i)t w

(i)tminus1η


(i)t |x

(i)1tminus1 u1tminus11113872 1113873

π x(i)1t|z1t u1tminus11113872 1113873

(3)



Power supplysystem

Motor controlmodule



WiFi

PC


(a) (b)




p xt|x(i)t m


p zt|xt m(i)


p zt|x(i)tminus1 m

(i) utminus11113872 1113873

(4)


w(i)t w

(i)tminus1η

p zt|xt m(i)


p x(i)t |x

(i)tminus1 m

(i) zt utminus11113872 1113873

propw(i)tminus1η

p zt|xt m(i)


p zt|x(i)tminus1 m

(i) utminus11113872 1113873

w(i)tminus1p zt|x

(i)tminus1 m

(i) utminus11113872 1113873

(5)


t


points

μ(i)t

1η(i)

1113944

K

j1xjp zt|xj m

(i)tminus11113872 1113873p xj|x


1113944(i)

t

1η(i)

1113944

K

j1p zt|xj m

(i)tminus11113872 1113873p xj|x



t1113872 1113873T

(6)


η(i) 1113944

K

j1xjp zt|xj m

(i)tminus11113872 1113873p xj|x

(i)tminus1 utminus11113872 1113873 (7)


Neff 1

1113936Ni1 1113957w

(i)1113872 1113873

2 (8)



below is solved


N

i11 minus M Si(ξ)( 11138571113858 1113859

2 (9)




sinψ cosψ1113890 1113891

six

siy

⎡⎣ ⎤⎦ +px

py

⎡⎣ ⎤⎦ (10)


Δξ Hminus 1

1113944

N


zSi(ξ)

zξ1113890 1113891

T

1 minus M Si(ξ)( 11138571113858 1113859 (11)



zξ1113890 1113891

T


zξ1113890 1113891 (12)






4 Path Planning







f(n) g(n) + h(n) (13)





1113868111386811138681113868 + y1 minus y21113868111386811138681113868

1113868111386811138681113868 (14)


Δx x + vΔt cos θt( 1113857 (15)

Δy y + vΔt sin θt( 1113857 (16)


xt+1 xt + vΔt cos θt( 1113857 (17)

yt+1 yt + vΔt sin θt( 1113857 (18)

θt+1 θt + ωΔt (19)









Vd (vω)|

2dis(vω) _vd

1113969

ge vc

2dis(vω))ωd

1113969

geωc1113882 1113883

(22)



















5m

5m

(a) (b)


1 2 3 4


1 2 3 4





1 2 3 4





(a) (b)


(a) (b)

Figure 10 Continued




(c) (d)








(a) (b)


(a) (b)



(a) (b)


(a) (b)


(a) (b)

Figure 17 Continued





6 Conclusions




(c) (d)



Data Availability




Acknowledgments


References


























p xt|x(i)t m


p zt|xt m(i)


p zt|x(i)tminus1 m

(i) utminus11113872 1113873

(4)


w(i)t w

(i)tminus1η

p zt|xt m(i)


p x(i)t |x

(i)tminus1 m

(i) zt utminus11113872 1113873

propw(i)tminus1η

p zt|xt m(i)


p zt|x(i)tminus1 m

(i) utminus11113872 1113873

w(i)tminus1p zt|x

(i)tminus1 m

(i) utminus11113872 1113873

(5)


t


points

μ(i)t

1η(i)

1113944

K

j1xjp zt|xj m

(i)tminus11113872 1113873p xj|x


1113944(i)

t

1η(i)

1113944

K

j1p zt|xj m

(i)tminus11113872 1113873p xj|x



t1113872 1113873T

(6)


η(i) 1113944

K

j1xjp zt|xj m

(i)tminus11113872 1113873p xj|x

(i)tminus1 utminus11113872 1113873 (7)


Neff 1

1113936Ni1 1113957w

(i)1113872 1113873

2 (8)



below is solved


N

i11 minus M Si(ξ)( 11138571113858 1113859

2 (9)




sinψ cosψ1113890 1113891

six

siy

⎡⎣ ⎤⎦ +px

py

⎡⎣ ⎤⎦ (10)


Δξ Hminus 1

1113944

N


zSi(ξ)

zξ1113890 1113891

T

1 minus M Si(ξ)( 11138571113858 1113859 (11)



zξ1113890 1113891

T


zξ1113890 1113891 (12)






4 Path Planning







f(n) g(n) + h(n) (13)





1113868111386811138681113868 + y1 minus y21113868111386811138681113868

1113868111386811138681113868 (14)


Δx x + vΔt cos θt( 1113857 (15)

Δy y + vΔt sin θt( 1113857 (16)


xt+1 xt + vΔt cos θt( 1113857 (17)

yt+1 yt + vΔt sin θt( 1113857 (18)

θt+1 θt + ωΔt (19)









Vd (vω)|

2dis(vω) _vd

1113969

ge vc

2dis(vω))ωd

1113969

geωc1113882 1113883

(22)



















5m

5m

(a) (b)


1 2 3 4


1 2 3 4





1 2 3 4





(a) (b)


(a) (b)

Figure 10 Continued




(c) (d)








(a) (b)


(a) (b)



(a) (b)


(a) (b)


(a) (b)

Figure 17 Continued





6 Conclusions




(c) (d)



Data Availability




Acknowledgments


References


























4 Path Planning







f(n) g(n) + h(n) (13)





1113868111386811138681113868 + y1 minus y21113868111386811138681113868

1113868111386811138681113868 (14)


Δx x + vΔt cos θt( 1113857 (15)

Δy y + vΔt sin θt( 1113857 (16)


xt+1 xt + vΔt cos θt( 1113857 (17)

yt+1 yt + vΔt sin θt( 1113857 (18)

θt+1 θt + ωΔt (19)









Vd (vω)|

2dis(vω) _vd

1113969

ge vc

2dis(vω))ωd

1113969

geωc1113882 1113883

(22)



















5m

5m

(a) (b)


1 2 3 4


1 2 3 4





1 2 3 4





(a) (b)


(a) (b)

Figure 10 Continued




(c) (d)








(a) (b)


(a) (b)



(a) (b)


(a) (b)


(a) (b)

Figure 17 Continued





6 Conclusions




(c) (d)



Data Availability




Acknowledgments


References



























Vd (vω)|

2dis(vω) _vd

1113969

ge vc

2dis(vω))ωd

1113969

geωc1113882 1113883

(22)



















5m

5m

(a) (b)


1 2 3 4


1 2 3 4





1 2 3 4





(a) (b)


(a) (b)

Figure 10 Continued




(c) (d)








(a) (b)


(a) (b)



(a) (b)


(a) (b)


(a) (b)

Figure 17 Continued





6 Conclusions




(c) (d)



Data Availability




Acknowledgments


References

























5m

5m

(a) (b)


1 2 3 4


1 2 3 4





1 2 3 4





(a) (b)


(a) (b)

Figure 10 Continued




(c) (d)








(a) (b)


(a) (b)



(a) (b)


(a) (b)


(a) (b)

Figure 17 Continued





6 Conclusions




(c) (d)



Data Availability




Acknowledgments


References



























1 2 3 4





(a) (b)


(a) (b)

Figure 10 Continued




(c) (d)








(a) (b)


(a) (b)



(a) (b)


(a) (b)


(a) (b)

Figure 17 Continued





6 Conclusions




(c) (d)



Data Availability




Acknowledgments


References


























(a) (b)


(a) (b)

Figure 10 Continued




(c) (d)








(a) (b)


(a) (b)



(a) (b)


(a) (b)


(a) (b)

Figure 17 Continued





6 Conclusions




(c) (d)



Data Availability




Acknowledgments


References



























(c) (d)








(a) (b)


(a) (b)



(a) (b)


(a) (b)


(a) (b)

Figure 17 Continued





6 Conclusions




(c) (d)



Data Availability




Acknowledgments


References




























(a) (b)


(a) (b)



(a) (b)


(a) (b)


(a) (b)

Figure 17 Continued





6 Conclusions




(c) (d)



Data Availability




Acknowledgments


References

























(a) (b)


(a) (b)


(a) (b)

Figure 17 Continued





6 Conclusions




(c) (d)



Data Availability




Acknowledgments


References




























6 Conclusions




(c) (d)



Data Availability




Acknowledgments


References

























Data Availability




Acknowledgments


References

























Documents

2D Lidar-Based SLAM and Path Planning for Indoor Rescue ...ruggedsurface,IMUisalsoinaccuratewhichmeansCar-tographermaygetbadresults.So,wechooseHector-SLAM. 4. Path Planning Toachievepathplanningistosolvethreebasicproblems: