Model parameter learning method and movement mode determination method

A movement mode determination method of a robot, performed to learn a movement of a first pedestrian with a plurality of second pedestrians surrounding the first pedestrian, where the first pedestrian and the plurality of second pedestrians are separate from the robot, and applying the movement that is learned to the robot, the movement mode determination method comprising:

• learning, by a learning device, a model parameter of a movement model;
• determining, by a movement mode determination device, a movement mode of the robot by using the movement mode model of which the model parameter is learned;
• determining a movement trajectory of the robot using the movement mode;
• and
• controlling the robot to move autonomously by controlling an actuator driving a movement mechanism of the robot, so that the robot follows the movement trajectory while traveling in an area including pedestrians,
• wherein the model parameter is learned by a model parameter learning method in which the learning device learns the model parameter of the movement mode model for determining the movement mode when the robot moves to a destination under a condition that a plurality of pedestrians are present around the robot,
• the model parameter learning method performed by the learning device, comprising:
• a first step of acquiring, in a case where the first pedestrian moves to the destination while avoiding interference with each of the plurality of second pedestrians under a condition that the plurality of second pedestrians are present around the first pedestrian, link data which are data in which a time series of a first movement mode parameter representing the movement mode of the first pedestrian in each movement period and a time series of a plurality of second movement mode parameters respectively representing the movement modes of the plurality of second pedestrians synchronized with the time series of the first movement mode parameter are linked to each other;
• a second step of extracting condition satisfying data which are link data satisfying a predetermined condition from the link data acquired in the first step, extracting feature amount vectors from the condition satisfying data, and linking the feature amount vectors to the condition satisfying data, wherein a condition that a movement trajectory of the first pedestrian does not have a portion whose curvature radius is a predetermined curvature radius or less is set as a primary condition, and satisfying the primary condition corresponds to satisfying the predetermined condition;
• a third step of determining, by a predetermined method, the number of clusters when clustering the feature amount vectors linked to the condition satisfying data in the second step;
• a fourth step of clustering, by a predetermined clustering method, the feature amount vectors of all clusters corresponding to the number of clusters determined in the third step;
• a fifth step of acquiring the condition satisfying data to which the feature amount vectors are linked as learning data in a case where the clustering of the feature amount vectors of all the clusters is completed;
• and
• a sixth step of learning the model parameter of the movement mode model by a predetermined machine learning algorithm using the learning data.