Projects for students (Internship, BTP, MTP, Course)

Student projects are divided in three categories. The expected outcome depends on the duration of the project and therefore, students are not supposed to complete the project. Instead, their performance will be evaluated on the basis of efforts they put in learning. In addition, my inputs will be proportional to the efforts of the student. The three categories of projects are

  1. Review and tutorial: You can select a few papers relevant as per your interest in which control and learning both are present. You can summarize those papers or their problem formulation, try to make a tree to connect other related papers and attempt to provide your own perspective. I will grade you on the basis of consistency, exhaustiveness, and completeness. You can write an article on medium or a post on LinkedIn to get noticed by the people or industries working in the related area. In this way, you will be able to secure your next career stop. Please do not use my name in your social media post without my permission.
  2. Reproduce and benchmarking: You can select a paper according to your interest and my recommendations, and reproduce their simulation results. If you have some novelty, you can make your git repository public. This open source contribution will help you to demonstrate your skills in front of interviewers and boost your resume. Please do not use my name anywhere in your git repository without my permission.
  3. Research and invent: For the course project, you can choose any topic within the general theme of the course for your research in a team of maximum four members. You can continue the work that you are doing with other faculty members with their permission. I will provide the publishable and applicable research directions to students who are working with me and willing to follow ethics policy.

You should consider the papers published in top AI and robotics conferences. You can choose papers written by famous people so that you can understand their research direction. You can also see the work carried out in top universities and industries.

Ongoing projects

Advanced RL for Quadrupeds robots developed for Police

This project focuses on developing advanced reinforcement learning (RL) algorithms for autonomous quadruped robots tailored for police operations. The robots are designed to navigate complex terrains, assist in surveillance, and support search-and-rescue missions. The integration of AI-driven decision-making enhances adaptability, efficiency, and real-time responsiveness in dynamic law enforcement scenarios.

Learning-based MPC for quadruped robots developed for police

This project develops learning-based Model Predictive Control (MPC) strategies for quadruped robots designed to assist police in real-world missions. The approach combines data-driven learning with predictive control to ensure agile, stable, and adaptive locomotion across unstructured environments. These robots enhance operational safety and efficiency in surveillance, search, and rescue tasks for law enforcement.

Reinforcement Fine Tuning of Large Language Model

This project focuses on Reinforcement Learning-based Fine-Tuning (RLFT) of Large Language Models (LLMs) to align model behavior with desired objectives. By integrating reward-driven optimization, the approach enhances performance, safety, and task-specific adaptability of LLMs. The project aims to improve model reliability and usefulness across diverse real-world applications.You should explore different methods of fine tuning of pre-trained LLMs (Deepseek, Gemma) including PPO and GRPO.

Remaining useful life estimation for an under water vehicle

This project focuses on estimating the Remaining Useful Life (RUL) of underwater vehicles using data-driven and physics-informed models. The approach enables early prediction of system degradation, ensuring mission reliability and reducing maintenance costs. Accurate RUL estimation enhances operational safety and extends the service life of autonomous underwater platforms.

Physics informed neural networks and scientific machine learning for Li-ion batteries

This project leverages Physics-Informed Neural Networks (PINNs) and other scientific ML approaches to model and predict the behavior of Li-ion batteries with improved accuracy and efficiency. By embedding electrochemical principles into neural network training, the approach ensures physically consistent and data-efficient learning. The framework aims to enhance battery management, diagnostics, and lifetime prediction in real-world applications.

Statistical analysis of spring faillures in LHB coaches of Indian railway

This project involves statistical analysis of spring failures in LHB coaches of the Indian Railways to identify key failure patterns and contributing factors. The study uses historical data and probabilistic models to assess reliability and predict failure trends. Insights from the analysis aim to improve maintenance strategies and enhance passenger safety.

Adaptive and intelligent perception system by integrating reinforcement learning with computer vision

This project explores the integration of reinforcement learning (RL) techniques in computer vision tasks to enable adaptive and intelligent perception systems. RL-driven models are trained to optimize decision-making in dynamic visual environments, improving performance in tasks like object detection, tracking, and scene understanding. The approach enhances the system's ability to learn from interactions and adapt to complex real-world scenarios.

Ethics and plagiarism