Davranış Temelli Robotlarda karmaşık karar verme süreci, bir giriş bilgisine belirli şekilde yanıt veren basit davranışlar üzerine kurulmuştur. Her bir davranış o anki giriş bilgilerine göre yanıtını belirler, sonra bu yanıtlar birleştirilerek elde edilen sonuç robotun kararı olarak uygulamaya konulur. Her biri ayrı güdü ile hareket eden davranışlar her zaman paralel istemlerde bulunmayabilir; birbirlerine zıt istemlerde bile bulunabilirler. Dolayısıyla robotun vereceği kararın başarısı, yanıtların ne şekilde birleştirileceği ile yakından ilişkilidir. Davranış tanımından veya karar verme kurallarından dolayı sürecin içerdiği belirsizlikler problemi zorlaştırmaktadır. Bu makalede, davranışların koordinasyonu için Belirsizlik Matrisi kavramı çerçevesinde yeni bir yöntem önerilmektedir. Bir davranışın tüm yanıtları Belirsizlik Matrisi ile gösterilip birleştirilerek davranışın ana yanıtı, ardından bu şekilde bulunan değerler bir araya getirilerek karar verme sisteminin ortak yanıtı elde edilmektedir. Yöntemin sınanması için; ‘hedefe doğru git’, ‘engelden kaçın’, ‘gezin’ ve ‘doğru git’ şeklinde dört davranışı bulunan bir mobil robot ele alınmış ve engellerin bulunduğu bir ortamda hedef noktasına gitme problemi için bilgisayar ortamında simülasyon düzeneği oluşturulmuştur. Bu çalışmada önerilen yöntem, davranış seçme matrisine göre belirli bir davranışın yanıtını ana yanıt olarak kabul eden klasik yöntem ile karşılaştırılmış ve önerilen metot açısından olumlu sonuçlar elde edilmiştir. Farklı yerleşim düzenleri ile yapılan denemelerde, önerilen yöntemin, robotu hedefe daha kısa sürede ulaştırdığı, yol süresince robotu engellerden daha uzakta tuttuğu ve ardı ardına gelen çelişkili adımları azalttığı görülmüştür.

Because robots are becoming more complex every day, the decision-making system which determines how the robot shall act, also gets equally complicated, becoming more difficult to manage. These difficulties may be overcome by grouping all movements and tendencies of the robot as ‘behaviors’. A behavior is a mechanism which describes the reaction and response corresponding to a stimulus. From this perspective, decision-making systems are revised into behaviors and consequently simplified. In this approach, the problems of which behavior to consider and how to combine responses to get an optimum result, are critical factors for attaining success. Response of a behavior is independent of the global purposes of the robot; it always gives the same reaction to the same action. Therefore it should be managed with a higher-level layer so that the robot could accomplish its global purposes. The responses of behaviors are converted into action by evaluation according to a coordination mechanism and then necessary action is taken. The success of the behavioral robot is closely related to the coordination system. The most widely used coordination methods are as follows; (1) Arbitration: The behavior with the highest priority will be executed, (2) Action- Selected: The selected action depending on the conditions and preferences present at the time is executed, (3) Voting: The response winning the majority of the votes is taken into account and the rest is disregarded, (4) Collective: The responses of the active behaviors are collected and the necessary action is executed. The process of combining behaviors involves various types of uncertainties including (1) inputs used to determine responses of the behaviors, (2) rules regarding the type of response given by the “behavior” to these inputs, (3) if more than one behavior is considered, the rules to determine which one of these should be selected as the “dominant behavior.” A tool is necessary to represent and process these uncertainties existing in the responses. Uncertainty matrix is able to fulfill this task. A new approach is proposed in this work that coordinates responses of all behaviors using Uncertainty Matrix. In this method, responses of each behavior are represented by Uncertainty Matrix and the combined response of the behavior is calculated. Then the main result is found with the same manner. The navigation problem of behavior-based robots to reach the target point while keeping away from the obstacles was simulated. The simulation environment consists of a 20 m × 30 m sized rectangular platform and one behavioral mobile robot placed in this, one unit of target point, three units of mobile or fixed obstacle blocks with a diameter of 1 m, and one unit of obstacle block with a diameter of 3 m. The robot has four behaviors, namely, move to goal, avoid obstacles, go ahead, and walk around. Two methods of behavior coordination are compared: (1) the classical approach that selects a behavior using behavior selection matrix and (2) the approach proposed in this work that coordinates responses of all behaviors using uncertainty matrix. In the second method, all responses of each behavior are represented by uncertainty matrix and the combined response is calculated. In cycles, 0.2 s each, the situation is reevaluated according to two methods and the robot moves on to the next point. In a typical starting position, the method proposed enables the robot to reach the goal in 61.2 s while the classical method reaches the goal in 78.2 s. The proposed method: (1) Makes the robot reach the goal in a shorter time, (2) Keeps the robot away from the obstacles. The method takes the nearest obstacle into account. Therefore the robot moves by keeping away from the obstacles, (3) Has less consequent contradictory steps, (4) Compared to classic methods, the resultant values obtained by using the uncertainty matrix are more meaningful. In this method, the probability value, that is “the certainty of selection of the angular value,” is also obtained besides the angular value. Positive results have been achieved in the simulation of the method proposed for the combination of the uncertainty, represented by uncertainty matrix, in the responses of the behaviors related to the determination of direction in mobile robots. Improvements at other stages of the decision-making processes of the behavior-based robots are possible by using the concept of the uncertainty matrix.