Environments and Tools Gazebo
Gazebo is 3D virtual model simulator for robot that operates by ROS.
Caffe
Caffe is one of the deep learning frameworks that is being developed as an open source.
Evaluation Methods Error Function
Represent the closeness measure of function with a training data.
・Cross Entropy
� = − ∑ ∑ ������ � �;
�
�=�
�=�
Stochastic gradient descent
Repeatedly update the weights and biases.
�� ≡ �= [ ��⋯ �] �
�+� = � − ��
�
Results and Evaluation Number of trials with loss function
Result model structure
Example Image Results
Future works and Conclusion.Classification Driving State using
Convolution Neural Network
Wasin Hawaree (福井工業大学),松永尭 (福井工業大学),信川創 (福井工業大学)
Background and Purpose
RoboCup Rescue Virtual Robot League (RCRL) in this competition to
perform the robot on a 3D space modeling in the disaster area and discovered
the location of victims. In this study, we consider driving state of the robot
with a convolution neural network.
Consider about classification of the complex images.
Furthermore, in obstacle not just the path and wall.
Study more about algorithm techniques.