MOBILE ROBOT NAVIGATION ON PARTIALLY KNOWN MAPS USING THE A* ALGORITHM

AQTR 2010 THETA 17th edition May 28-30 2010 Cluj-Napoca Romania, Proceedings of 2010 IEEE International Conference on Automation, Quality and Testing, Robotics. http://aqtr.ro, Paul I. Muntean
Abstract In these paper I will present a path planning solution for mobile robots navigation based on the A * algorithm which is one of the most used algorithms in path planning now days . For routing requirement under three dimensions is much more complex than under two dimensions, the traditional A* algorithm should be improved to meet the routing requirement. In these paper the traditional A* algorithm will be used since the maps used for navigation are 2D maps. IDA* is a space-efficient version of A*, but it suffers from cycles in the search space (the price for using no storage), repeated visits to states (the overhead of iterative deepening), and a simplistic left to-right traversal of the search tree.

I. INTRODUCTION Recent studies are showing that every day activity of people in cities and countries in modern society is rapidly rising so there is a need for navigating peoples movement . Researchers have tried to come with new and better navigation technologies in the last years[3](Jones, 2001). In order to understand path finding it is necessary to have a short look on the history . All started in the nineteen seventies when some scientists started research on the routing solution for moving chess in the chessboard or moving fragment in the puzzle map[2](Eklund et al., 1996). The main reason for starting the research in these fields is that these problems can be easily abstracted and further on the results can be applied to more complex fields of study. And with the development of path finding, several new classical routing algorithms have been introduced to generate better routing solution. For example the Dijkstra algorithm is the most famous one, which evaluates the moving cost from one node to any other node and sets the shortest moving cost as the connecting cost of two nodes[2](Eklund et al., 1996). Around at the same period of time Best-First-Search algorithm is also introduced in the researchers community. A little different from the Dijkstra algorithm, Best-First-Search algorithm has an different approach because it estimates the distance from current position to goal position , and it chooses the step that is more closer to the goal position[1](Amit). The difficulty was growing with the new path finding situations so the old path finding algorithm had to be improved to meet the new introduced requirements. A new path finding algorithm was introduced and it was named the A* algorithm.The A* algorithm tries to combine the advantages offerd by Dijkstra algorithm and Best-First-Search algorithm, A* algorithm tries for each movement to take the shortest step and it also takes into account about the choosing step whether on the direction which is just from source to target[3](Jones, 2001).

The key point of research has become algorithms efficiency in the moment when the real development of the A* algorithm started. The A* algorithm is a breadth first algorithm, it uses huge amounts of memory to keep the data of current proceeding nodes[5](Nelson and Toptsis, 1992). During the traversing of all nodes which are possible to be placed on the optimized path the disadvantage is that it uses a huge size of stack which is needed to contain the considering grids. Not only the development of the A* algorithms and its own efficiency are for interest , new methods of using A* algorithm are also considered by the researchers. Like the bidirectional A* algorithm searching method which was used in order to reduce the time cost of the A* algorithm. The most important difference of the bidirectional A* algorithm compared to the classical A* algorithm which is searching from source to target is that as his name says that it searches from source to target and vice versa. The searching stops immediately when the two directions searching progresses meet each other in bidirectional A* algorithm[5](Nelson and Toptsis, 1992). Now days regarding the computer society it can be observed an three dimensional trend , three dimensional A* algorithms development has caught more attentions. The A* algorithm can be used also with some small differences in computing paths along 3D maps which are more like the real world challenges that path planning algorithm will have to face .An example would be the path planning of a cart in a mine system which has more levels. One frequently used approach for solving three dimensional path planning problem, is to map the three dimensional map into a two dimensional expression in order to use traditional A* algorithm for solving the path finding request[4](Makanae and Takaki, 2004). These technique of mapping 3D to 2D is working for path finding requirement under some simple 3D situations, under complex situations these mapping method could not easily be used for path finding so the recommendation is to use a more advance method. Lets take for example the restricted spatial situations such as mining and inner of tall buildings where the overlapping layers may appear frequently, and these situations are impossible to be solved by using the traditional A* algorithm solution, for mapping 3D into 2D and deriving the optimum path are nearly impossible under these special circumstances. The A* algorithm can be improved to meet these new challenging situations. As an evolution of the A* algorithm the three dimensional A* algorithm emerged which can work with more restrictive situations. A few precise modifications should be taken for the

standard A* algorithm and the result is the new improved 3D A* algorithm.

2.THE A* ALGORITHM The [12]A* algorithm uses a best-first search and finds the least-cost path from a given initial node to one goal node (out of one or more possible goals). It uses a distance-plus-cost heuristic function (usually denoted f(x)) to determine the order in which the search visits nodes in the tree. The distance-plus-cost heuristic is a sum of two functions.The path-cost function, which is the cost from the starting node to the current node (usually denoted g(x)) and an admissible "heuristic estimate" of the distance to the goal (usually denoted h(x)).

these neighbors are added to the queue. The algorithm continues until a goal node has a lower f value than any node in the queue (or until the queue is empty). (Goal nodes may be passed over multiple times if there remain other nodes with lower f values, as they may lead to a shorter path to a goal.) The f value of the goal is then the length of the shortest path, since h at the goal is zero in an admissible heuristic. If the actual shortest path is desired, the algorithm may also update each neighbor with its immediate predecessor in the best path found so far; this information can then be used to reconstruct the path by working backwards from the goal node. Additionally, if the heuristic is monotonic (or consistent,), a closed set of nodes already traversed may be used to make the search more efficient. 3.IMPLEMENTATION

F= G+ H

(1)

The h(x) part of the f(x) function must be an admissible heuristic; that is, it must not overestimate the distance to the goal. Thus, for an application like routing, h(x) might represent the straight-line distance to the goal, since that is physically the smallest possible distance between any two points or nodes. If the heuristic h satisfies the additional condition

h(x)d(x,y)+h(y)

(2)

for every edge x, y of the graph (where d denotes the length of that edge), then h is called monotone, or consistent. In such a case, A* can be implemented more efficiently, roughly speaking, no node needs to be processed more than once, and A* is equivalent to running Dijkstra's algorithm with the reduced cost . d'(x,y): = d(x,y) h(x) + h(y) (3) As A* traverses the graph, it follows a path of the lowest known path, keeping a sorted priority queue of alternate path segments along the way. If, at any point, a segment of the path being traversed has a higher cost than another encountered path segment, it abandons the higher-cost path segment and traverses the lower-cost path segment instead. This process continues until the goal is reached. Like all informed search algorithms, it first searches the routes that appear to be most likely to lead towards the goal. What sets A* apart from a greedy best-first search is that it also takes the distance already traveled into account; the g(x) part of the heuristic is the cost from the start, not simply the local cost from the previously expanded node. Starting with the initial node, it maintains a priority queue of nodes to be traversed, known as the open set (not to be confused with open sets in topology). The lower f(x) for a given node x, the higher its priority. At each step of the algorithm, the node with the lowest f(x) value is removed from the queue, the f and h values of its neighbors are updated accordingly, and

The Open List in A* is implemented as a balanced binary tree sorted on f values, with tie-breaking in favor of higher g values. This tie-breaking mechanism results in the goal state being found on average earlier in the last f-value pass. In addition to the standard Open/Closed Lists, marker arrays are used for answering (in constant time) whether astate is in the Open or Closed List. We use a lazy-clearing scheme to avoid having to clear the marker arrays at the beginning of each search. Each path finding search is assigned a unique (increasing) id that is then used to label array entries relevant for the current search. The above optimizations provide an order of magnitude performance improvement over a standard textbook A* implementation[6]. function A*(start,goal) closedset := the empty set// Set of nodes already evaluated. openset := set containing the initial node // Tentative nodes to be evaluated. g_score[start] := 0 // D. from start along optimal path. h_score[start] := heuristic estimate distance(start, goal) f_score[start] := h_score[start]// Estimated total distance from start to goal through y. while openset is not empty x := the node in openset having the lowest f_score[]value if x = goal return reconstruct_path(came_from[goal]) remove x from openset add x to closedset foreach y in neighbor_nodes(x) if y in closedset continue tentative_g_score := g_score[x] + dist_between(x,y) if y not in openset Figure 1.The A* algorithm part 1

add y to openset tentative_is_better := true elseif tentative_g_score < g_score[y] tentative_is_better := true else tentative_is_better := false if tentative_is_better = true came_from[y] := x g_score[y] := tentative_g_score h_score[y] := heuristic_estimate_of_distance(y, goal) f_score[y] := g_score[y] + h_score[y] return failure function reconstruct_path(current_node) if came_from[current_node] is set p = reconstruct_path(came_from[current_node]) return (p + current_node) else return current_node

Fig 3.Map used for Algorithm testing After applying the A* algorithm under the same condition by changing his parameters it was need a comparison between the classical A* algorithm and a more advanced A* algorithm which is faster if we because it leaves the open node in the priority queue. T.N. 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 H 0 1 2 3 4 0 1 2 3 4 5 6 7 8 0 1 2 3 4 0 1 2 3 4 5 0 1 2 D y y y y y y y y y y y y y y y y y y y y y y y y y y y y F. P. y y y y y y y y y y y y y y y y y y y y y y y y y y y y F. m m m m m M(x,y) M(x,y) M(x,y) M(x,y) M(x,y) M(x,y) M(x,y) M(x,y) M(x,y) D. S. D. S. D. S. D. S. D. S. E E E E E E SQR SQR SQR T [sec] 0.0975 0.0302 0.0104 0.0033 0.0034 0.1025 0.0494 0.0275 0.0173 0.0110 0.0108 0.0087 0.0075 0.0078 0.108 0.0286 0.0114 0.0034 0.0037 0.1120 0.0499 0.0254 0.0205 0.0146 0.0153 0.1235 0.0155 0.0012

Figure 2.The A* algorithm part 2 The closed set can be omitted (yielding a tree search algorithm) if a solution is guaranteed to exist, or if the algorithm is adapted so that new nodes are added to the open set only if they have a lower f value than at any previous iteration. 4.EXPERIMENTS These section is divided into two sections. In the first section I will present test results obtained with a application designed for testing the A* algorithm under different environment condition and also when using different parameters under the same conditions map conditions. The second part will present the test results obtained with another application designed for mobile robot path planning in an partially known environment[9]. The goal of these second part is not only to observe the test results and times obtained for different runs but also to see hove the robot control by the algorithm remotely manages to follow the path and avoids unknown obstacles. 4.1 The A* vs. Fast A* algorithm test results In order to determine the time needed to determine an optimum path from the starting position to the target position the application uses a high resolution timer which was implemented using the Kernel32.dl , the obtained times obtained for calculating the path have 4 decimals so it can be guaranteed that the test results are precise and.

29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46

3 4 0 1 2 3 0 1 2 0 1 2 3 0 1 2 0 1

y y y y y y y y y y y y y y y y y y

y y no no no no no no no no no no no no no no no no

SQR SQR m m m m M(x,y) M(x,y) M(x,y) D. S. D. S. D. S. D. S. E E E SQR SQR

0.0011 0.0015 30 1.3443 0.0124 0.0322 10.685 2.7441 30 30 1.3152 0.0358 0.3453 25.305 2.2041 30 25.966 30

action added to the robot which has a distance parameter set for slow mode to 225 which represent millimeters and for fast mode it is set to 50 millimeters, actually these tells the robot how to interpret the data obtained from the ultrasonic sensors. The obstacles ar represented by the boxes with two diagonal lines inside them.It can be observed that the obstacles that appear in the maps loaded in the Pioneer simulator are not present in the PathPlanner application so the robot has to avoid them in order to go to his final destination.For each of these two maps the time will be determined.

Table 1.Test Results for A* vs A* Fast The Table 1 must be explained:T.N. (test number),H (heuristic),D(allow yes/no diagonals),F.P.(yes/no A* Fast),F(formula),T(Time),y(yes),n(no),m(manhattan),M(x,y) (Max(Dx,Dy)),D.S(diagonalshortcut),E(euclidean),SQR(Euclid ean without square).First test run the fast A* algorithm and for each formula the heuristic number was increased until the time need for path calculation started to rise ,this was the moment when the next formula was taken and the proces started again.The second test run the exact procedure was applied but with the classical A* algorithm. 4.2 Path Planning with the Pioneer robot. Testing with the Pionner 2 robot Simulator[7] robot using two unkwon environments. The PathPlaning application used for testing the Pionner Simulator was eveloped for handling partially unknown maps like the two that will be further on preseneted.For these simulation I used the C12 laboratory map.One time with 1 unknown obstacle and the second time with 2 unknown obstacles each time unknown for the simulator.These two obstacels were only present in the Pioneer Map because the scope is to simulate a real environment where the robot has to sense the world and based on sensor readings to decide the further necessary actions, like for example avoiding obstacles. The map used in the path planning application is a 2D map composed of squares wich in real life are of dimension 10x10 centimeters.The application can handle maps up to 10x10 meters.The path planning solution was completly developed in java[8]programming language using the JNI (Java Native Interface) ,the [11]Aria library .In order to simulate the two maps the runs were needed for each of the two maps because two different running modes were tested .The first one is called fast mode and the second one slow mode[10].Fast mode differes from slow mode in some action limiters that are not added to the robot and also by chanching the parameter of a Figure 3.The Path Planner map.

Fig 4.The First Pionner Simulator Map

The Table 2. Shows that when ussing the Fast mode variant of the A* algorithm on the Second Map the time decreases with 5 seconds wich is not to be neglegted. When simulating the same algorithm on the first map wich has two obstacles the time increases because the range of the sonars is reduced to 50 millimeters and so the obstacle is detected later which ends in a recover actions needed to recover the robot and so precios time is lost. Since these application uses only data collected from ultrasonic sensors it is recomanded to use sensor fusion and to adapt the application to real applications which use a combination of radar, GPS and laser sensors in order to offer to the mobile robot a more precise description of the environment.

RERENCES

Fig 5.The Second Pionner Simulator Map Test run 1 2 3 4 Fast Mode Yes No Yes No Map S.Map S.Map F.Map F.Map Time [sec] 40 45 61 47

Table 2. Test results for the path planner 5.CONCLUSIONS The data obtained from the A* algorithm simulator shows a big difference between the two variants of the A* algorithms. The Fast A* algorithm allows to increase the heuristic number for seven up to eight runs insteed the classic A* lets the an increase up to value of three . The shortest time is also obtained for the Fast A* algorithm wich also shows that the performances of the classical A* algorithm can be further increased in order to gain more speed. It is recomanded for further research on these topic to use more advanced variant of the A* algoritm like the bidirectional A*, the IDA* algorithm or D* algoritm. Regarding the results obtained from the simulation of the A* algorithm on the Pionner Simulator the tests should be made on more type of maps wich is not a problem since the Path Planner application supports creation and importing of maps and so does the Pioneer Simulator application.

[1] Amit, Amit's Game Programming Information. http://wwwcsstudents. stanford.edu/~amitp/gameprog.html. (accessed 28 Jan.2008) [2] Eklund, P.W., Kirkby, S. and Pollitt, S., 1996. A Dynamic Multi-source Dijkstra' Algorithm for Vehicle Routing. In: N.a. Jain (Editor), Conf. on Intelligent Information Systems, Australian New Zealand. [3] Jones, J.H., A* Tutorial. http://www.geocities.com/jheyesjones/astar.html. (accessed 28 Jan.2008) [4] Makanae, K. and Takaki, M., 2004. Development of the 3Dimensional Urban Spatial Data Model and Application to the Pedestrian Navigation System. ITS . [5] Nelson, P.C. and Toptsis, A.A., 1992. Unidirectional and Bidirectional Search Algorithms. Digital Object Identifier, 9(2): [6]. Siegwart R., Nourbakhsh Illah R., Introduction to Autonomous Mobile Robots, Massachusetts Institute of Technology, 2004 [7]. Saphira Operations and Programming Manual Version 6.2, MobileRobotsROBOTICS, 1999 [8]. Konolige, K., K. Myers The SAPHIRA architecture: a design for autonomy. In Artificial Intelligence Based Mobile Robots: Case Studies of Successful Robot Systems, MIT Press, 1998.[10] WiBox 2100 Quick Start Guide, Lantronix, 2004 [9] F. Lingelbach, Path Planing for mobile manipulation using probabilistic cell decomposition.Inteligent Robots and Systems, 2004 (IROS 2004).Proceedings ,IEEE,vol3,pages 2807-2812,2004. [10] J.Rosel and P.Iniguez Path planning using Harmonic Functions and Probabilistic Cell Decomposition, Procedings of the IEEE International Conference on Robotics and Automation ,pages 1803-1808, [11] ActivMedia Robotics, Inc. Aria Reference Manual 1.1.10, November 2002. [12] The A* algorithm .http://en.wikipedia.org/wiki/A*_search_algorithm