Outlining my non-published projects here (competitions, course research projects, designed homeworks/assignments). My research work can be seen here.

Tech Competitions

Micro Aerial Vehicle Swarm Formation Challenge sponsored by Drona Aviation : This problem statement to make a trajectory of a swarm of drones by creating a ros-independent pipeline. The micro aerial vehicles were localised using markers and an external vision system, where each communicated self position to the ground station. The control was completely centralised. We used a deadlock system pipeline for the drones to move in a synchronous motion otherwise a failsafe for maintaing current position would trigger.
Github Link: aerial/interiit11 | Problem Statement: PS | Solution : report

Multi-Payload Delivery Challenge using UAV sponsored by Flipkart GRiD 4.0 : This problem statement required to prepare a pickup-drop package pipeline for drones used in rescue operations. With the takeoff, the drone would grid search a area to look for any packages to be picked up (which would be identified using vision) and grabbed using an autonomous triggering electromagnet. The drone would then ascend back to its hovering height to drop the package to a drop-off location and continue the grid search operation from where it left off.
Github Link: aerial/shastra23 | Problem Statement: PS

Other Research (Course Projects)

Autonomous Landing of UAV using Model Predictive Controller
Course: CS637 (Embedded and Cyber-Physical Systems): This project is a ROS simulation of a IRIS UAV (by eth-z) landing on a the top a husky platform. A MPC controller has been setup for predicting the optimal trajectory according to the reference trajectory fed to the drone. Two cases have been tested; one where the platform is stationary and other where the platform is moving. This is done inorder to test the response and accuracy of the controller. However due to issues with velocity contoller, the simulation is not present on our github link. Please stay tuned for updates on this.
Github Link: rahul/mpc_autoland

Machine Unlearning using model treatment techniques
Course: CS772 (Probabilistic Machine Learning): This project is exploring probabilistic approaches for treating model performance. This is required for safety critical tasks where an example that the model was previously trained on needs to be removed and so the model needs to completely forget about it. The naive method suggests to train the model from scratch on the new dataset; however, machine unlearning explores methods to avoid extensive retraining and smartly subtract the influence of the unrequired examplesfrom the trained parameters. For this, we devised a two-stage pipeline; 1) Cause identification for identifying what datapoints in the dataset are faulty (or no longer required) and 2) Model treatment to use a mathematical formula to rectify the trained weights. Achieved scores of 0.42 for corrupted training data and 0.50 for filtered training data after treatment.
Github Link: rahul/pml | Project Report: Solution