cv

Ph.D. in Computer Science (4.3/5.0)

• Thesis: A framework for learning and evaluating Social Robot Navigation (Advisor: Prof. Harold Soh)
• Expected October 2026

M.S. in Robotics and Autonomous Systems: AI Concentration (4.29/5.0)

• Thesis: Learning interpretable action models of simulated agents through agent interrogation

B.Tech in Mechanical Engineering (3.88/5.0)

• Thesis: Geometrical Mapping of an Initially Unknown Region by a Mobile Robot

Ph.D. Researcher (Embodied AI)

• Language-Driven Scenario Generation for Social Navigation
  • Developed an LLM-powered pipeline that translates natural language descriptions into executable human-robot interactive testing scenarios, reducing scenario synthesis time to ~1 minute, and compared three planners in a 160-participant case study. [Under Review: IEEE RAL] (Paper, Code)
• Largest Expert Demonstration Social Navigation Dataset
  • Co-led an eight-team open-source social navigation dataset collection effort (29.5 hours of onboard robot data from 7 robots across 5 countries, 43.5 hours BEV data). Independently analyzed multimodal onboard sensor data and benchmarked foundational models. [Under Review: IJRR] (Website)
• Learning Residual Social Corrections for Navigation Policies
  • Designed and trained sample-efficient end-to-end and residual imitation-learning social navigation policies using abstract representations to reduce the sim-to-real gap, iterating on architectures and training paradigms through scenario-based testing. (Ongoing Work)

Research Assistant

• Conducted practical experiments to test an AI action-model learning algorithm in game-based environments. (Paper)
• Engineered a robot demo using ROS, OpenRAVE, AutoCAD, and Fusion 360 to demonstrate robots assisting humans in the assembly of automotive parts (NSF-Funded Project). (Demo)

Robotics Intern

• Implemented a LiDAR sensor fusion ROS2 package for autonomous forklifts and deployed it in simulation using Docker.

Research Intern

• Generated a ROS/Gazebo dataset and trained TensorFlow models to approximate a robotic-arm inverse-kinematics solver, achieving 98% test accuracy. Investigated RNNs and LSTMs for motion planning. (Code)

Research Intern

• Simulated a decentralized multi-robot graph-based exploration algorithm using Gazebo and ROS, and achieved complete coverage of random maze environments. (Code)

• Best-Paper Runner-Up, RSS 2024 Workshop on Unsolved Problems in Social Navigation

• First team from NITK to qualify for the national round of ABU-Robocon

Object pose hallucination for object search under full occlusion

• Investigated 6-DoF fully-occluded object pose estimation in cluttered scenes with diffusion models, and a differentiable-rendering loss-guidance signal that steers sampling towards observation-consistent poses. (Code)

Explanation Generation in Human-Robot Teaming

• Implemented a framework that uses inverse reinforcement learning to learn the preferred order of explanations provided by an AI agent to humans, to minimize the cognitive load in collaborative tasks. Designed human-study experiments (Amazon MTurk) to evaluate the proposed algorithm. (Code)

ABU-Robocon NITK

• Designed, developed, prototyped, fabricated, and assembled (as a team) two robots that play a cooperative game of shuttlecock throwing; became the university's first team to qualify for the national round. (Link)

Learning robotic snake locomotion using Genetic Algorithms

• Created a framework that uses genetic algorithms to mimic snake motion from a video onto a snake robot simulated in CoppeliaSim. (Code)

• Datasets for Social Navigation, Embodied AI Workshop, National Robotics Programme
• Curating Social Navigation Datasets for Training Navigation Models, NSCC-NRP Workshop on Advancing Embodied AI Research through Data

• Towards Automated Scenario-based Testing of Social Navigation Algorithms, RSS 2024 Workshop on Unsolved Problems in Social Navigation (Best-Paper Runner-Up)