My work and research is in the field of Artificial Intelligence and Computer Vision for Robotics(mostly Autonomous Vehicles). Working in this domain I have gained knowledge in various sub-fields of this domain and also gathered skills which help me formulate and model solutions for the various problems like localization, state-estimation and decision making for the autonomous agent. I am also familiar with various frameworks like Keras, TensorFlow, OpenCV, ROS, and hardware like Arduino, RaspberryPi, ARM-CortexM.
The motivation and driving factor for me is the opportunity to learn new concepts(state-of-the-art) and build solutions to problems and also innovate. Nevertheless it an exciting, interesting and ever-evolving field to work-in which also provides for great research opportunity.
One of the most complex problems I have solved is in the field of Autonomous Navigation for Self-Driving Cars. For this I had to build various models to detect the drivable area, lane-markings and also other vehicles on the road. It was also important for this model to run in real-time. After this Computer Vision aspect was completed the next step was to work on the lane-keeping of the car (i.e localize the car within the lane using the lane markings on the road) by using the error generated if the vehicle deviated to much from its lane. This was not the complete solution but there are many cases where the algorithm could struggle, like if there are on lane-markings. It was a good start to a great challenge, and seeing the vehicle move autonomously was a great feeling.
Another project I am proud of is the use of Spiking Neural Networks(SNN's) which are a third generation of neural networks for the classification task. This was a great project because it a more closer implementation of the working of the human brain similar to the action-potentials generated in the neurons of the brain. These networks are also computationally more efficient if the right hardware is developed.
At the end of the day I am an individual passionate about technology and driven by the opportunity to innovate and research.
Worked with Object-Detection Algorithms like YOLOv3, for performing Early Prediction and Detection by reducing the overall inference time.
Clustering problems on a million face data-set for Face Recognition using various techniques like DDC(Deep Density Clustering) and UMAP(Uniform Manifold Approximation and Projection for Dimension Reduction)
Played a key role in this start-up in end-to-end Lane Segmentation for Drivable-Space and Lane-Marking Detection for an Autonomous Vehicle.
Worked on the technique of Behaviour Cloning for a Golf-Cart’s Motion Planning and Navigation.
Also worked on R&D of Neural Network Optimization Techniques like Weight Pruning and Weight Sharing for the networks to be implemented on Embedded Systems.