CN114832372B - High-efficiency simulation marble AI simulation method based on evaluation function - Google Patents

Abstract

The invention discloses an efficient simulation marble AI simulation method based on an evaluation function. By adopting the marble AI simulation method provided by the invention, the problem of AI simulation requirements under the conditions that a strategy result cannot be estimated and a space is infinitely solved in a complex process can be solved. The invention greatly saves the planning engineering quantity of developers for the AI process, and can flexibly regulate and control the evaluation function for the demand of simulation AI so as to lead the AI to show different behavior modes, thereby being more simulation and intelligent. Meanwhile, compared with the traditional three schemes, the technology is lower in development and maintenance cost. Furthermore, the technology has very high expansibility, and besides physical ejection games, the scheme can be suitable for AI requirements with complex and uncontrollable processes and particularly large solving space.

Description

High-efficiency simulation marble AI simulation method based on evaluation function

Technical Field

The invention relates to the technical field of game production engine software and development systems, in particular to a high-efficiency simulation marble AI simulation method based on an evaluation function.

Background

AI techniques currently used in network games are generally as follows:

FSM (Finite State Machine)

Generally, the method is used for simple AI, uses different states and state switching to describe and control the AI behavior, and has the disadvantages of difficult realization and maintenance and limited use scene for complex requirements.

Behavior Tree behavior Tree

The basic principle is to abstract the logic required to control AI into different node types, including behavior nodes, condition nodes, combination nodes, modification nodes, etc., and then organize these nodes in a tree, and when an AI decision is needed, traverse the tree from top to bottom until a eligible leaf node (behavior) is found and then execute it.

Compared with a state machine, the behavior tree has the advantages of clearer logic, better abstract packaging, more suitability for more complex AI requirements and current mainstream AI implementation mode. The disadvantage is that the performance burden is heavy due to the principle of traversing the tree, and meanwhile, after the AI requirement scale is enlarged, the complexity of the behavior tree is still very high, and the maintenance cost is also high.

GOAP

For the demand of further complexity, the complexity of the behavior tree also rises sharply, and for some more flexible application scenarios, the GOAP is a more suitable AI decision manner. The feature of GOAP is to let the AI decision system make the best solution by itself, given the goals and behaviors, rather than limiting the way state switches as in behavior trees. Meanwhile, each behavior changes the state of the world, so that the method has certain adaptivity.

Compared with a state machine and a behavior tree, the GOAP scheme relieves the energy of developers from the switching planning of states, decouples the targets and the behaviors, and has higher flexibility in AI. However, GOAP has its disadvantages, and planning by GOAP is performed at runtime, which has a certain impact on performance. Meanwhile, controllability is relatively weak, and AI behaviors outside the planning are easy to occur.

The three methods are mainstream AI implementation methods at present, and are generally used in a situation where a solution space is limited, and when the number of behavior choices of the AI is too large and a result is difficult to plan, the above-mentioned schemes cannot be well applied. Taking a pinball game project as an example, if a battle AI is to be made, the AI needs to select one transmitting direction from 360-degree directions, and a collision result after transmission is physically simulated, so that the collision result is difficult to estimate, and the situation is complicated and changeable. Since the transmit directions are continuous, the solution space is equal to infinity and cannot be planned across possibilities. None of the three conventional schemes described above achieve a very effective and efficient immersive AI.

Thus, the prior art is deficient and needs improvement.

Disclosure of Invention

The invention aims to overcome the defects of the prior art and provides an efficient simulation marble AI simulation method based on an evaluation function.

The technical scheme of the invention is as follows: the high-efficiency simulation marble AI simulation method based on the evaluation function comprises the following steps:

step 1: simulating a battlefield, and carrying out AI simulation operation;

step 2: after the AI simulation operation is finished, respectively calculating the score condition of single operation according to each numerical value generated by the operation and the weight of the numerical value by taking the result as the guide through an evaluation function;

and step 3: according to the score condition obtained in the step 2, calculating the simulation degree of the simulation, including the behavior characteristics which are planned by the developer and are close to the real operation and the behavior characteristics which are not similar to the real operation, and scoring the characteristic behaviors;

and 4, step 4: calculating or comparing and analyzing according to the weight to obtain a final score, namely the optimal solution of the simulation AI;

and 5: and applying the obtained optimal solution to AI actual operation.

Further, the specific steps of step 1 are:

step 1.1: preparing N AI battlefields, and putting into simulation operation;

step 1.2: when the AI needs to make a decision, copying the current battlefield situation into the AI battlefield;

step 1.3: the AI selects n directions from 360 degrees directions to carry out operation execution;

step 1.4: the AI manager records various data in the execution process as the calculation basis of a subsequent evaluation function;

step 1.5: and when all units of the AI battlefield stop, ending the simulation and entering the next stage.

Further, the specific steps of calculating the score in step 2 are as follows:

step 2.1: taking the current nth execution operation, including a collision score, a skill score or a reply score, and respectively calculating the score condition;

step 2.2: the nth collision score is Collide (n) = collision injury = offensive power of injured ball;

step 2.3: the nth Skill score Skill (n) = Skill effect value = effect weight × the force of attack by effect ball;

step 2.4: the nth recovery score is Heal (n) = treatment amount = treatment weight × subject's offensive power.

Further, the step 3 of simulating truth and scoring specifically comprises the following steps:

step 3.1: taking the current as the nth time of adding score, and then scoring the nth time of fidelity degree

Simulate (n) = (colloid (n) × beat addition + Skill (n) × Skill effect addition × beat addition + Heal (n)) × nth scoring weight;

step 3.2: current direction this simulation total score is total (n) = Simulate (1) + Simulate (2) +. + Simulate (n).

By adopting the scheme, the Marble AI simulation method provided by the invention can solve the problem of AI simulation requirements under the conditions that a strategy result cannot be estimated and a space is not solved infinitely in a complex process. The invention greatly saves the planning engineering quantity of developers for the AI process, and can flexibly regulate and control the evaluation function for the demand of simulation AI so as to lead the AI to show different behavior modes, thereby being more simulation and intelligent. Meanwhile, compared with the traditional three schemes, the technology is lower in development and maintenance cost. Furthermore, the technology has very high expansibility, and besides physical ejection games, the scheme can be suitable for AI requirements with complex and uncontrollable processes and particularly large solving space.

Detailed Description

The present invention will be described in detail with reference to specific examples.

The invention provides an evaluation function-based high-efficiency simulation marble AI simulation method, which comprises the following steps:

step 1: and simulating a battlefield and carrying out AI simulation operation. The method comprises the following specific steps:

step 1.1: preparing N AI battlefields, and putting into simulation operation;

step 1.2: when the AI needs to make a decision, copying the current battlefield situation into the AI battlefield;

step 1.3: the AI selects n directions from 360 degrees directions to carry out operation execution;

step 1.4: the AI manager records various data in the execution process as the calculation basis of a subsequent evaluation function;

step 1.5: and when all units of the AI battlefield stop, ending the simulation and entering the next stage.

First, N AI battlefields are prepared in advance, and when the decision is needed in the turn of AI rounds, the current actual battlefield situation is copied into the AI battlefield. The AI then selects n directions from the 360 degree directions for transmission. The simulation speed after transmission can be different from that of formal combat, so as to accelerate the simulation time. During the simulation, the AI manager records all important data, such as the number of collisions, the number of collisions at the edge of the battlefield, etc., as a basis for the calculation of the subsequent evaluation function. And when the simulation of all units in the AI battlefield is finished, entering the next stage. The AI battlefield number N and the direction number N may be adjusted freely according to the specific situation such as performance, and the larger N or N is, the more accurate the AI result is, and the smaller N or N is, the less performance consumption is.

Step 2: after the AI simulation operation is finished, the evaluation function takes the result as the guide, and the score condition of the single operation is respectively calculated according to each numerical value generated by the operation and the weight of the numerical value. The specific steps for calculating the score condition are as follows:

step 2.1: taking the current nth execution operation, including a collision score, a skill score or a reply score, and respectively calculating the score condition;

step 2.2: the nth collision score is Collide (n) = collision injury = offensive power of injured ball;

step 2.3: the nth Skill score Skill (n) = Skill effect value = effect weight × the force of attack by effect ball;

step 2.4: the nth recovery score is Heal (n) = treatment amount = treatment weight × subject's offensive power.

And step 3: and (3) calculating the simulation degree of the simulation according to the score obtained in the step (2), wherein the simulation degree comprises the behavior characteristics which are planned by the developer and approximate to the real operation and the behavior characteristics which are not planned by the real operation, and the characteristic behaviors are scored. The simulation scoring specifically comprises the following steps:

step 3.1: taking the current as the nth time, and then the nth simulation degree is scored

Simulate (n) = (colloid (n) × beat addition + Skill (n) × Skill effect addition × beat addition + Heal (n)) × nth scoring weight;

step 3.2: current direction this simulation total score is total (n) = Simulate (1) + Simulate (2) +. + Simulate (n).

And 4, step 4: and calculating or comparing and analyzing according to the weight to obtain a final score, namely the optimal solution of the simulation AI.

And 5: and applying the obtained optimal solution to the AI actual operation.

After the simulation is finished, the evaluation function is calculated according to the situation in the simulation, the direction of the optimal solution is selected, and the method is applied to AI actual operation. The evaluation function is composed of two parts, namely a result-oriented evaluation function part, the actual benefits are concerned, such as total damage to an enemy caused by the operation, total blood volume returned by the enemy, skill effect and the like, and each value has different weights so as to reflect the influence degree of the different values on the final win. And the other part of the evaluation function calculates the simulation degree of the simulation, comprises the behavior characteristics which are planned by the developer and approximate to the real operation and are not similar to the real operation, scores the characteristic behaviors, and finally calculates the final score by a weight to obtain the optimal solution of the simulation AI. The method comprises the following specific steps:

for the resulting oriented evaluation function, assume that the current is the nth collision, the nth collision score Collide (n) = collision injury × force of attack of injured ball, the nth Skill score Skill (n) = Skill effect value × effect weight = force of attack of affected ball, and the nth recovery blood volume score Heal (n) = treatment volume = treatment weight = force of attack of treated person.

In addition to the weight plus the unit attack force, the unit attack force acts as a scale for the unit strength and the degree of importance, and thus each benefit is multiplied by the unit attack force to carry the weight embodiment of AI for the target selection.

The next step is the plausibility evaluation function, which scores Simulate (n) = (Collide (n) × beat addition + Skill (n) × Skill effect addition × beat addition + Heal (n)) × the nth additive weight, assuming that it is the nth additive. When there is no beat condition, the beat is added to =1; when there is a beat condition, the beat is added to = the beat weight for the hero.

The implication of multiplying the immersive function by the nth scoring weight is that if there is unreasonable wall-strike rebound and hitting an enemy after hitting a teammate, which is not a real operation, then the AI is not favored for this choice because as n increases, the scoring weight decreases, and the overall score decreases.

Finally, the current direction simulates this time the total score = Simulate (1) + Simulate (2) +.. Simulate (n). (n = number of collision scores)

And respectively comparing the total scores of the N directions, wherein the direction with the highest score is the optimal solution of the operation of the most simulation player.

In summary, the marbles AI simulation method provided by the invention can solve the AI simulation requirement problem under the conditions that the strategy result cannot be estimated and the space is infinitely solved in the complex process. The invention greatly saves the planning engineering quantity of developers for the AI process, and can flexibly regulate and control the evaluation function for the demand of simulation AI, so that the AI can show different behavior modes and is more simulation and intelligent. Meanwhile, compared with the traditional three schemes, the technology is lower in development and maintenance cost. Furthermore, the technology has very high expansibility, and besides physical ejection games, the scheme can be suitable for AI requirements with complex and uncontrollable processes and particularly large solving space.

The present invention is not limited to the above preferred embodiments, and any modifications, equivalent substitutions and improvements made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims (1)

1. An efficient simulation pinball AI simulation method based on an evaluation function is characterized by comprising the following steps:

step 1: simulating a battlefield, and carrying out AI simulation operation; the method comprises the following specific steps:

step 1.1: preparing N AI battlefields, and putting into simulation operation;

step 1.2: when the AI needs to make a decision, copying the current battlefield situation into the AI battlefield;

step 1.3: the AI selects n directions from 360 degrees directions to carry out operation execution;

step 1.4: the AI manager records various data in the execution process as the calculation basis of a subsequent evaluation function;

step 1.5: when all units of the AI battlefield stop, after the simulation is finished, entering the next stage;

step 2: after the AI simulation operation is finished, respectively calculating the score condition of single operation according to each numerical value generated by the operation and the weight of the numerical value by taking the result as the guide through an evaluation function; the specific steps for calculating the score condition are as follows:

step 2.1: taking the current nth execution operation, including a collision score, a skill score or a reply score, and respectively calculating the score condition;

step 2.2: the nth collision score is Collide (n) = collision injury = offensive power of injured ball;

step 2.3: the nth Skill score Skill (n) = Skill effect value = effect weight × the force of attack by effect ball;

step 2.4: the nth recovery blood volume score is Heal (n) = treatment volume treatment weight for the offensive power of the person to be treated;

and step 3: according to the score condition obtained in the step 2, calculating the simulation degree of the simulation, including the behavior characteristics which are planned by the developer and are close to the real operation and the behavior characteristics which are not similar to the real operation, and scoring the characteristic behaviors; the simulation scoring specifically comprises the following steps:

step 3.1: taking the current as the nth time of the bonus, the nth time of the plausibility degree is scored as Simulte (n) = (colloid (n) × beat addition + Skill (n) × Skill effect addition + Heal (n)) × nth time of the bonus weight;

step 3.2: current direction this simulation total score is total (n) = Simulate (1) + Simulate (2) +. + Simulate (n);

and 4, step 4: calculating or comparing and analyzing according to the weight to obtain a final score, namely the optimal solution of the simulation AI;

and 5: and applying the obtained optimal solution to the AI actual operation.