RESEARCH AND PUBLICATIONS
I am, among other things, interested in Machine Learning, Robotics and Language Grounding. I currently study problems that arise when training sensorimotor policies for robotics on tasks that do not afford an easily characterizable system state representation, let alone it's estimation. Specifically, by transferring learned policies from simulation to real robots.
Christoforos I. Mavrogiannis, Valts Blukis and Ross A. Knepper, Socially Competent Navigation Planning by Deep Learning of Multi-Agent Path Topologies, IEEE/RSJ International Conference on Intelligence Robots and Systems (IROS), Vancouver, BC, CA, September 2017. [PDF]
PAST ENGINEERING PROJECTS
Waiterbot (1st Runner-up, Tech Factor Challenge 2014-2015)
Waiterbot is a hand-built prototype for a fully featured restaurant robot. We built it in a team of 4 enthusiasts (Azfer Karim, Jagapriyan, Mohamed Musadhiq and me) for the 2014-2015 Tech Factor Challenge competition in Singapore. It can navigate from table to table, pick up and deliver orders and even serve water. It uses a custom-built gripper to pick up cups, bowls and plates (although it struggles with crockery).
Battlefield Extraction Robot (Grand Prize, Tech Factor Challenge 2013-2014)
ABER (Autonomous Battlefield Extraction Robot) is a rescue robot that can automatically find and pick up a casualty in a war or disaster scenario, where limited remote operation is possible. It's built on ROS and uses a unique conveyor-belt pickup system that inflicts minimum damage in case of bone fractures. This project earned me the DSTA Gold Medal for the best final-year project in NTU, School of Electrical and Electronic Engineering.
Investigating Heatmaps of Adversarially Perturbed Images (project for CS6784)
In this class project I thought it would be fun to perform Layer-wise Relevance Propagation for deep neural networks (paper) to generate heatmaps for images that have been adversarially perturbed. Adversarial perturbations, such as the ones generated by DeepFool (paper) perform tiny, imperceptible adjustments to the input image to force a deep neural network classifier, in this case a ResNet-51, to misclassify the image as another class with high confidence. They can be generated quite trivially by gradient descent on the input pixels. I found that in many cases, perturbations significantly shift the focus of the network, such as in the below (highly cherry-picked, but interesting) example, where forcing a cab to be classified as a car wheel has drawn focus on the actual car wheel. Spatial focus shift occured in roughly half the perturbed images.