Publications & Projects

For a complete list of publications, please visit my Google Scholar profile: Google Scholar. A few recent projects are listed below:

Text2Traffic: Retrieval Enhanced In-Context Learning For Complex Air Traffic Scenario Generation

Text2Traffic methodology diagram

Text2Traffic Methodology: Based on user input, Text2Traffic leverages Retrieval-Augmented Generation (RAG) to extract relevant sector information, including airways, waypoints, and aircraft types, from a domain-specific document, enabling the generation of the requested air traffic scenario.

Tools and technologies used
Retrieval Augmented Generation (RAG) LLMs (GPT3.5-Turbo, Llama3.1-8b-Instruct) Langchain Syntactic and semantic performance evaluation Maximal Marginal relevance (MMR)

Towards conformal automation in air traffic control: Learning conflict resolution strategies through behavior cloning

Conformal automation diagram

End-to-end conflict resolution pipeline using ATCO-conformal ML models.

Tools and technologies used
Supervised machine learning (Random forest, XGBoost, SVMs) Human-in-the-loop experiments Data analytics and Visualization Model robustness analysis

An Agent-Based Approach for Air Traffic Conflict Resolution in a Flow-Centric Airspace

Flow-centric conflict resolution

A concept diagram for the interaction between the agent and the learning environment. Scenarios involving interflow and intra-flow conflict are generated and the vector representation of the extracted features is used by the agent to propose an action based on the learned policy, thereby reaching a new state and receiving a certain reward. The updated actions pass into a self-stabilizing algorithm which ensures intra-flow safe separation and outputs the updated location and speed of the aircraft in both flows.

Tools and technologies used
Deep reinforcement learning (PPO, DQN) Agent-based modeling

Continuous descent flight operations in Singapore Changi airport

Tools and technologies used
Random forest regression Feature engineering