1-(608) 504 8032
Academic Projects
Developing a new machine learning approach using CNNs and SNNs for a 2 DOF robotic manipulator control task
Master of Science Thesis
In this task, a new machine learning method was proposed by integrating Convolutional neural networks and Spiking neural networks for the task of target recognition and obstacle avoidance conducted by the robotic platform.The networks were organized in a consecutive order. A Faster R-CNN detector was trained using the VGG16 neural network followed by a proposed algorithm to align the CNN output to the SNN. The algorithm was capable of expanding neural regions of the SNN. (Paper Submitted to NeuroComputing journal-Elsevier).
A video is attached below. For more information about the project, please contact me.
Integrating Auto-encoders and SNNs to be implemented on a service robot
The project is currently in progress. The main objective of this project is to use the encoded section of Auto-encoders in order to encode the input to be used as the input data for the sensory neural region in Spiking neural networks. A deep autoencoder has been modeled and trained using the Tensorflow-Keras library by 2000 augmented pictures and the encoder section has been extracted from the network. The extracted encoder section is imported in Matlab to connect the encoder output to the input layer of the SNN. (Article is under preparation)
For more information about the project, please contact me.
Localization and optimal path planning of mobile robots in unknown environments by integrating CNNs and Optimal Control
The project is currently in progress. The contribution of this research is to devise a local map using the CNNs and MLPs for the optimal control on a mobile robot for the task of optimal decision making and path planning in the absence of GPS devices. The mobile robot is designed and built and a Yolo V2 detector is currently on the training phase using 1000 augmented images. An algorithm is written using the Matlab Programming Language to calculate the costs of different places in the picture so that these costs can be later used by the optimal control unit for decision making.
For more information about the procedure of the project, please contact me.