| CPC G05D 1/0088 (2013.01) [G01S 17/89 (2013.01); G01S 17/931 (2020.01); G05D 1/024 (2013.01); G05D 1/0274 (2013.01)] | 6 Claims |

|
1. A mobile robot for evaluating self-training based traversability comprising:
an elevation map generator which generates a grid-cell based elevation map using point cloud data;
a feature extractor which extracts a plurality of types of features on each grid cell from the elevation map;
a data set generator which generates a labeled data set which is labeled and an unlabeled data set for training, based on labeling features set for at least two types of features among the plurality of types of features; and
a self-training unit which generates an AI model for evaluating traversability by self-training using the labeled data set and the unlabeled data set,
wherein the data set generator generates the labeled data set in such a manner that a grid cell having the labeling features above a preregistered threshold is decided to be non-traversable and the features of the corresponding grid cell are labeled as negative, and that a grid cell having the labeling features below the preregistered threshold is decided to be traversable and the features of the corresponding grid cell are labeled as positive,
wherein the grid cell labeled as positive is extracted only on the navigation trajectory which is the path on which the mobile robot already traveled, among the grid cells having the labeling features below the preregistered threshold, and
wherein the features for the other grid cell not included in the labeled data set is generated as unlabeled data set,
wherein the self-training unit is configured to perform:
training a classifier using the labeled data set as training data;
predicting traversability for the grid cell consisting of the unlabeled data set by inputting the unlabeled data set as input data to the trained classifier;
classifying the grid cell consisting of the unlabeled data set into traversable grid cell, non-traversable grid cell and unlabeled grid cell, based on the prediction results;
labeling the features of the traversable grid cell as positive, labeling the features of the non-traversable grid cell as negative, and then adding the labeled features to the labeled data set; and
recreating the unlabeled data set by means of the features of the unlabeled grid cell;
wherein the trained classifier is created as the AI model by repeating the training a classifier to the recreating the unlabeled data until a predetermined termination requirement is met.
|