MUHAMMAD HUSEIN ABDUL HALIM. Modification of Navigation System on
Assistant Harvesting Robot using Deep Learning. Supervised by SLAMET WIDODO.
Greenhouse farming is often a suitable field for applying the technologies of
automation, computing and robotics. The robots can perform some tasks that humans cannot do due to the harsh conditions of greenhouses, such as environmental monitoring and control, crop monitoring, supply and treatment, and pest and disease detection. The prototype robot from the previous studies has some lack on the movements of robot that are still difficult to turn and need color objects for do object tracking. This has disadvantages because the color objects can be interference with another objects of the same color which can affect the accuracy of object tracking. Another method that can be used for detecting object is deep learning. Deep learning is a subset of machine learning in artificial intelligence (AI) that has networks capable of learning unsupervised from data that is unstructured or unlabeled. This research aims to modificate the navigation system on assistant harvesting robot using deep learning. The deep learning model to be used is Convolutional Neural Network. This research will be carried out in several stages: 1) Problem identification, 2) Formulate the problem and solution, 3) Design analysis, 4) Build Prototype, 5) Test and evaluate the prototipe. For an object localization problem the image used as an input, which goes through a Convolutional Neural Network that results in a vector to classify the object. To localize the object in the image, the neural network need a few more output units that encompass a bounding box. The output is four more numbers, which identify the x and y coordinates of the upper left corner and the height and width of the box. The coordinates are the basis for determining the movement of the robot according to the algorithm for the movement of the robot. Materials that will be used in the research are camera, DC motor and driver, Python programming language with openCV and Tensorflow. The expected result of this research is the prototype of assistant robot that can load and deliver harvested products in a greenhouse using a deep learning-based object tracking algorithm navigation system.
Key words: Assisting robot, deep learning, object tracking, harvesting, navigation system