When the codes are executed as mentioned above, the Kobuki was able to rotate, take multiple images and pass them to modified Inception v3 engine. Figure below shows the Kobuki rotating from its initial position, taking images at fixed intervals.
Kobuki Turtlebot2 scanning the area environment with the vision sensor.
Figure below shows the terminal in which Inception v3 engine using TensorFlow framework identifies different objects from the image. In the terminal, the identified object along with its accuracy (score) can be seen.
Terminal window displaying output of Inception v3 engine. The displayed output shows a backpack and a broom were discovered in the last two images.
The engine classifies the image and the object recognized were registered in a Object Position Discovery (OPD) file. The yaw angle from the initial position of the robot to each object was written in same OPD file against the recognized object label.
Sample of a generated Object Position Discovery file containing each recognized object and the corresponding angles to the objects.
The output of the ImageIdentifier script after cleaning up the csv file using FileCleanup script. All recognized objects were listed and ask the user to select the required object.
Terminal displaying the output of the IdentifyLocation script. The displayed output shows a prompt to the user asking for the name of the discovered object.
The robot orients itself towards the required object once it gets an input from the user. At this point the depthimage_to_laserscan package was activated. The distance of the object is calculated from this data.
ROS sensor visualization output displaying the PointCloud image taken from the Asus Xtion RGB-D camera.
Top view of simulated laser scan (/scan ROS topic) from Asus Xtion.
After finding the distance to move, the robot starts moving towards the object. Figures below shows the movement of robot taken at various instances. The Kobuki will stop in front of the object keeping a safe distance.
Kobuki Turtlebot2 robot moving towards the bucket object.
The rqt_graph of all ROS nodes are given below:
