JPH11102295A - Real-time search in combination with full search - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a real-time search with which the waste of energy can be reduced while reducing useless moving and the dynamic change of a search space can be followed up as well by using both a full search and the real-time search. SOLUTION: A full search A* and a real-time search RTA* are parallelly started from a start spot START at the same time and a search corresponding to one step 15 performed by both A* and RTA*. In this case, when a distance H from a spot B to be reached by A* for one step to a target spot GOAL is shorter than a distance D from a spot A reached by RTA* to the target spot GOAL, the search corresponding to one step is continued from the spot B to be reached by A* by both A* and RTA* but when the distance D is shorter than the distance H, the search corresponding to one step is continued from the spot A reached by RTA* by both A* and RTA*.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a real time search method combined with a full search method for searching for an optimum route from a start point to a destination point by using both a full search method and a real time search method.
For example, an autonomous mobile robot that interacts with a human in the real world, an interface agent system that performs an autonomous interaction between the Internet and a user, etc. are effective in searching for an action sequence to achieve a purpose given by the user. The present invention relates to a real-time search method combined with a search method.

[0002]

2. Description of the Related Art The search method is a technique for finding an optimum route from a certain initial state to a target state. For example,
Considering an autonomous mobile robot that distributes mails and the like in an office or the like, suppose that the robot at Mr. A is instructed to deliver a document to Mr. B. At this time, the robot has to search for an optimal movement route from the point A in the initial state to the point B in the target state.

[0003] To explain some terms here, the world searched by the search method is called a search space, and the actor performing the search is called a problem solver.
What should be noted here is the relationship between the problem solver and the robot. For example, in the example of the above-mentioned office robot, (1) the robot stores a map of the office in its own memory, searches for a route to the destination using the map in the memory, and actually searches the route according to the result. If the user moves, the search space in this case is a map of the office stored in the memory, and the problem solver is a program that has performed a search in the memory. On the other hand, in the case where (2) the robot does not have a map of the office in the memory and searches for the destination while actually moving in the office, the search space is the actual office and the problem solver is Be a robot. Hereinafter, in order to prevent confusion between the problem solver and the robot in the present specification, the problem solver will be used only in the meaning of the above (1), and the robot in (2) will not be called the problem solver.

[0004] The full search method is a search method used in the form of the above (1), and is a method of blindly searching the entire search space to find a "shortest path" to a target state, and is a classical method. A vertical search and a horizontal search are mentioned as examples. In either case, the shortest route to the destination can be obtained, but it takes an enormous amount of time to search, and is unsuitable as a search method for autonomous mobile robots and interface agents targeted by us. A heuristic search has been proposed as the same full search method. In the heuristic search represented by A ^* , the distance between the problem solver and the destination is estimated using an evaluation function, and this is used for the search, thereby dramatically reducing the search time compared to the classical method. (The Manhattan distance is famous as an evaluation function.)

[0005] However, it is difficult to apply A ^* even for the following reasons. Heuristic search works effectively if the accuracy of the evaluation function is accurate.However, if the structure of the search space and the purpose given are complicated, it becomes difficult to provide an appropriate evaluation function.
Since it takes the same search time as the classical method, it is also difficult to apply it to the robots we are targeting. Moreover, in the whole search method in general,
Since the robot starts moving along the route obtained after all the searches have been completed, there is also a disadvantage that if the search space changes dynamically during the search, it cannot follow the change. The full search method including A ^* is described in Artificial Intelligence Dictionary (Maruzen Publishing, 1991), pp. 754-758.

On the other hand, RTA ^* (Real-Time A ^* )
Is a search method proposed to solve the shortcomings of the full search method, and is a search method used in the form of the above (2). The real-time search is a search method that pursues only that the robot reaches the destination in a shorter time, regardless of finding the shortest route. More specifically, the robot arrives at the destination while alternately repeating a search step that always ends in a fixed short time and a step in which the robot moves immediately based on the result. If one search time is shorter than the change interval of the search space, it is possible to follow a dynamic change of the search space. As described above, the real-time search method has an advantage that the search can be performed in an extremely short time and that it can follow a dynamic change in the search space as compared with the full search method.

[0007] However, when applied to a robot or the like that is our target, the point that "it is not always possible to move to the target point by the shortest route" is a major obstacle for the following reasons. In the real-time search method, similarly to the above-described heuristic search method, the estimated distance to the destination is calculated using an evaluation function, and the search is performed using the calculated distance.
At this time, if the accuracy of the evaluation function is high, there is no problem, but if the search space and the given purpose become complicated, the accuracy of the evaluation function will decrease, and the robot cannot always reach the target point efficiently and a huge number of robots There is a disadvantage in that a meaningless amount of movement is performed. In the autonomous mobile robots and interface agents we are targeting,
This useless movement is a major obstacle. This is because useless movement in the robot causes unnecessary waste of battery and, in the interface agent, wasteful use of network resources. For the real-time search method including RTA ^* , "
Artificial Intelligence ”(Elsevier Science Publis
hers, 1990) pp. 189-211.

[0008]

As described above, A ^*
In the full search method, if the structure of the search space and the given purpose are complicated, it is difficult to provide an appropriate evaluation function.
In addition to the problem that it takes a long time to search, and because the full search method starts moving after completing all search processing, if the search space changes dynamically during the search,
There is a problem that it cannot follow this.

Similarly, in the RTA ^* search method, if the search space and the given purpose are similarly complicated, the accuracy of the evaluation function is reduced, and it is not always possible to efficiently reach the target point, and an enormous amount of action is required. There is a problem of performing meaningful movement and wasting energy.

In a situation where an autonomous mobile robot or an interface agent performs some kind of search, it is meaningless that only the shortest distance is guaranteed or only the shortest path is guaranteed. Must be obtained. Therefore, there is a need for a real-time search method that combines the features of both the full-search method and the real-time search method.

[0011] The present invention has been made in view of the above,
The goal is to use both the full search method and the real-time search method to reduce wasteful movement and waste of energy, and to follow dynamic changes in the search space. It is to provide a combined real-time search method.

[0012]

According to a first aspect of the present invention, there is provided a method for searching for an optimum route from a start point to a destination point using both a full search method and a real time search method. A real-time search method combined with a search method, in which the full search method and the real-time search method are simultaneously executed in parallel,
A search corresponding to one step is performed by both the full search method and the real-time search method, and a first distance from a point that can be reached by the full search method to a destination point by the search corresponding to the one step is determined by A second distance from a point reached by a search corresponding to one step by a real-time search method to a destination point is compared. If the first distance is shorter than a second distance, the full search method is used. The search corresponding to the one step is continued by both the full search method and the real-time search method from the reachable point, and when the second distance is shorter than the first distance, the real-time search method The search corresponding to the one step is continued from the reached point by both the full search method and the real-time search method, and when the point reached by the full search method or the real-time search method matches the destination point, the search processing is performed. The Exit The gist of the Rukoto.

According to the first aspect of the present invention, the full search method and the real time search method are simultaneously started in parallel, and the search corresponding to one step is performed by both the full search method and the real time search method. And the first distance from the point that can be reached by the full search method to the destination point by one step search is 1
When the search for the step is shorter than the second distance from the point reached by the real-time search method to the destination, one step is performed by the full search method and the real-time search method from a point reachable by the full search method. Is continued, and when the second distance is shorter than the first distance, the search corresponding to the one step is performed by both the full search method and the real time search method from the point reached by the real time search method. In order to continue
When the distance is shorter, useless movement by the real-time search method can be suppressed, and when the second distance is shorter, the search time by the full search method can be reduced.

[0014]

Embodiments of the present invention will be described below with reference to the drawings.

[0015] Figure 1 is a diagram for explaining a full-search method combined real-time search method according to an embodiment of the present invention, the robot A ^* full search method from a start point START and RTA
^* Real-time search is started in parallel at the same time, and a search corresponding to one step, which is a relatively short predetermined time, is performed. RTA ^* Point A and destination GOAL reached by the real-time search D and A between the distance ^* and the distance H between the point A and the destination GOAL which can be reached by the full search method. The distances D and H are compared, and the search method which obtains a smaller distance is obtained. This shows the process of performing the next search with the reached point as the next start point START.

First, as shown in FIG. 1A, at time t ₀
The robot that is the main actor at the start point STA
The shortest route search from RT to the destination GOAL is performed by A ^*.
TA ^* Search and movement are repeatedly started by the real-time search method. Then, at time t ₁ , the robot actually travels the distance d by the RTA ^* real-time search method and reaches the point A. In the search by the A ^* full search method, the estimated shortest path h to the point B is obtained. Can be The estimated distance from the point B, which is the tip of the estimated shortest route h obtained by the A ^* full search method, to the destination GOAL is obtained as H by the evaluation function, and the point A where the robot moves by the RTA ^* real time search method. The estimated distance from to the destination GOAL is obtained as D by the estimation function. In FIG. 1, a hatched area S is a search space searched by the A ^* all search method.

Here, a search speed which is an index of the progress of the search in the present embodiment will be described. The search speed is the average of the speed at which the estimated distance between the problem solver and the robot and the destination GOAL decreases until the search is completed, and is defined as _{VA *} and _{VRTA *} , respectively. For example, if the estimated distance between the start point START and the destination point GOAL is D
As a result of searching the search space of ist by the A * full search method, assuming that tA _* time is required, _{VA *} = Dist / tA _* . As a result of searching the same search space by RTA *, _assuming that t _{RTA *} time is required, V _{RTA *} = Dist / t
_{RTA *} . Note that “A *” added to the subscript,
“RTA *” means “A *” and “RTA *”, respectively (hereinafter the same).

The basic idea of the search method is based on the following two items. (1) When different search methods having the same search speed are executed at the same time, the shorter the distance estimated by the evaluation function between the problem solver and the robot and the destination, the shorter the possibility of completing the search first. (2) The search speed varies depending on the characteristics of the search method itself, but greatly changes during the search due to the structure of the search space, the given purpose, and the like.

Therefore, a new search method which can obtain a quasi-shortest path in a quasi-shortest time by using A * as a full search method and RTA * as a real-time search method in accordance with the following basic policy is considered. be able to.

(1) Policy 1: If DH> α (α is a threshold value), the robot moves from point A to point B and returns to RT.
A * is started, and A * also continues searching from point B.

(2) Policy 2: Conversely, if HD> β (β is a threshold value), point A is newly set as a start point START.
A * is started, the search process up to that point is discarded, and the robot continuously performs RTA * from point A.

The movement of the robot from the point A to the point B in the policy 1 returns to the start point START using the trajectory moved by the RTA *, and reaches the point B by the route obtained by the A *. With this method, the user can move from point A to point B without using a search.

Next, it is presumed that the optimum values of the thresholds α and β are different depending on the standard search speeds of the two search methods, and α and β are functions f () and g with arguments of V _{A *} and V _{RTA *} , respectively. Replace with () and change as follows.

(1) Policy 1: In policy 1, it is necessary to consider the distance h + d that the robot moves from point A to point B. Therefore, policy 1 is changed as follows.

[0025]

D> f ( _{VA *} , _{VRTA *} ) (h + d) + H (α = f
( _{VA *} , _{VRTA *} ) (h + d)), the robot moves from point A to point B and returns to RT again.
Execute A ^* .

(2) Policy 2: In policy 2, since the robot does not need to move as in policy 1, the robot is changed as follows.

[0027]

## EQU00002 ## If H> g ( _{VA *} , _{VRTA *} ) + D (.beta. = G ( _{VA *} , _{VRTA *} )), the START point is updated to the point A, and A ^* is executed again.

Then, when the speed ratio of V _{A *} and V _{RTA *} is changed using a plurality of prepared search spaces, f (), g when the robot can reach the destination in the shortest moving step.
(1) As a result of measuring the values, policies 1 and 2 are finally determined as follows.

[0029]

(Equation 3) If so, the robot moves from point A to point B and RT again
Execute A ^* (note that the γ and δ values depend on the search algorithm used).

[0030]

(Equation 4) If so, the start point START is changed to point A and A * is executed again (note that the ε, γ, and δ values depend on the search algorithm used).

Returning to FIG. 1A, at time t ₁ ,
When the estimated distance between the point B and the destination GOAL is obtained as H in the A * full search method, and the estimated distance between the point A and the destination GOAL is obtained as D in the RTA * real-time search method,
Comparing both distances H and D, it is found that the distance H is shorter than the distance D. Therefore, the above-mentioned policy 1 is applied, and the robot is moved from the point A to the point B as shown in FIG. RTA
* Search is restarted by real-time search method and A *
The full search method also continues the search from point B. In this case, the movement of the robot from the point A to the point B is performed by the method described above.

The search is continuously restarted from the state shown in FIG. 1 (b), and after a lapse of a predetermined time, the state shown in FIG. 1 (c) is reached. In this state, since the distance D is shorter than the distance H, the policy 2 is applied, the A * full search method is started with the point A as a new start point START, and the search process up to that point is discarded. The robot continuously executes the RTA * real-time search method from the point B. Then, after a lapse of a predetermined time, the search operation of reaching the state shown in FIG. 1D is repeated, and any one of them reaches the destination GOAL first, and the search processing ends.

As described above, in the present embodiment in which the search is continued while applying the policy 1 or the policy 2 based on the magnitude relationship between the distances D and H, for example, in a situation where the RTA * is performing a useless operation, It is highly probable that A * precedes and policy 1 is applied by D 適用 H, and as a result, useless operation, which is a problem of RTA *, can be suppressed.
Then, it is expected that the search by RTA * precedes, and as a result, D≪H, and policy 2 is applied.
A * starts searching from a new point by applying the policy 2, so that the searching time by A * can be reduced. As a result, since A * is executed intermittently, it is possible to follow a dynamic search space change to some extent.
Since the two search algorithms are executed to overtake each other in this way, Trade-off Search (TO
S).

Next, the operation of the real-time search method combined with the full search method of the present embodiment will be described with reference to the flowchart shown in FIG.

When the search is started, the RTA ^* real-time search method and the A ^* full search method are executed in parallel. First, both the RTA ^* and A ^* search methods are initialized (step S1).
1, S12), then both search methods start point STA
The search is simultaneously performed from the RT, and the search for one step is performed by the RTA ^* real-time search method and the A ^* full search method (steps S13 and S23).

The search for one step by the RTA ^* real-time search method is completed in a short time as described above.
Means that the robot moves once by searching twice,
A ^{* In the} full search method, a search is performed by regarding the search space as a graph space. Basically, the search is performed in the following steps. The problem solver is the starting point STAR at first
Although it is at T, it searches for a point that can be moved by one node from there, and moves to a point where the estimated distance to the destination GOAL is the shortest. Further, a point that can be moved by one node from that point is searched for, and movement is performed in the same manner.
Until this problem solver moves by one node, it is defined as a search for one step.

Next, in step S31, RTA ^*
The real-time search method determines whether the robot has reached the destination GOAL or the A ^* full search method determines whether the problem solver has found a route to the destination GOAL. If so, the search process ends. (Step S4
1) If not, it is determined whether or not the policy 1 is established at this time (step S33).
When the policy 1 is established, the robot moves from the point A to the point B, that is, at this time, to the point at the tip of the search by the A ^* all search method (step S35). And
Returning to steps S13 and S23, a new RTA ^* real-time search method is started from this point B (step S13), and a search using the A ^* full search method is also performed continuously from point B (step S23).

If policy 1 is not established, it is determined whether policy 2 is established (step S37). If policy 2 is established, point A is set to a new destination point GO.
The A ^* full search method is started as AL, and the previous search process by the A ^* full search method is discarded and initialized (step S39), and the process returns to steps S13 and S23.
While restarting the A ^* full search method from the point, the robot continuously starts the RTA ^* real-time search method from the point A.

If it is determined in step S37 that the policy 2 is not satisfied, the process returns to steps S13 and S23 to continue the search using the RTA ^* real-time search method and the A ^* full search method.

The following Table 1 shows the search time to the destination GOAL and the number of moving steps of the problem solver by the real-time search method combined with the full search method (shown as TOS in Table 1) of the present embodiment. Case and A ^* Full search and RT
A ^* shows simulation results of comparative evaluation in the case where each of the real-time search methods is independently searched. Note that the optimum f () and g () values are used as the search speeds _{VA *} and _{VRTA *} .

[0041]

[Table 1] As shown in Table 1, in comparison between the full search method and the real-time search method using the full search method of the present embodiment and A ^* full search method, A
^* The search time of the full search method was reduced, and the problem solver was able to reach the destination GOAL in 13% of the search time on average, compared to the case where the search was performed using the A ^* full search method alone. .

In comparison between the real-time search method using the full search method and the RTA ^* real-time search method according to the present embodiment, in particular, policy 1
, The number of useless moving steps in the RTA ^* real-time search method executed in this embodiment is reduced, and the RTA ^*
Compared to the case where the search is performed only by the real-time search method, the problem solver uses the number of moving steps of 19% on average, and
I was able to reach AL.

From the above simulation results, it was confirmed that the real-time search method combined with the full search method according to the present embodiment is capable of reducing unnecessary search while maintaining real-time performance.

[0044]

As described above, according to the present invention,
The full search method and the real-time search method are simultaneously executed in parallel, and a search corresponding to one step is performed by both the full search method and the real-time search method. When the first distance from the point to the destination to the destination is shorter than the second distance from the point to the destination by the search for one step from the point reached by the real-time search, the total distance from the point reachable by the full search is The search corresponding to one step is continued by both the search method and the real-time search method,
When the second distance is shorter than the first distance, the search corresponding to the one step is continued by the full search method and the real time search method from the point reached by the real time search method. When the distance is shorter, useless movement by the real-time search method can be suppressed, and thus waste of energy can be reduced. When the second distance is shorter, the search time by the full search method is reduced. And can follow dynamic changes in the search space.

According to the present invention, in the real world,
For example, the efficiency of a search performed by an autonomous robot that assists a person in an office or the like or an interface agent that autonomously mediates with a person on the Internet is improved. In addition, since the useless motion, which is a problem of the real-time search method, can be reduced, it is possible to reduce the waste of energy due to useless movement when applied to a robot, and to uselessly when applied to a software agent or the like. Waste of network resources can be reduced.

[Brief description of the drawings]

FIG. 1 is a diagram for explaining a real-time search method combined with a full search method according to an embodiment of the present invention.

FIG. 2 is a flowchart showing the operation of the real-time search method combined with the full search method shown in FIG.

Claims (1)

[Claims] A real-time search method combined with a full-search method and a real-time search method for searching for an optimal route from a start point to a destination point using a full-search method and a real-time search method. The execution starts in parallel and simultaneously, and the search corresponding to one step is performed by both the full search method and the real-time search method. From the point that can be reached by the full search method by the search corresponding to the one step, the destination point And a second distance from a point reached by a real-time search method to a destination point by a search corresponding to the one step, wherein the first distance is greater than a second distance. If it is shorter, the search corresponding to the one step is continued by both the full search method and the real-time search method from the point that can be reached by the full search method, and the second distance is shorter than the first distance If the real time When the search corresponding to the one step is continued from the point reached by the search method by both the full search method and the real time search method, and the point reached by the full search method or the real time search method matches the destination point And a real-time search method combined with a full search method, which ends the search processing.