Research & Projects

Research

Robotic Connector Manipulation - IIWA 14 Robot with NIST Board

Advisor: Dr. S K Gupta, RRoS Lab | Spring/2025

Built simulation environment in NVIDIA Isaac Sim 4.5 for IIWA 14 to perform plug/unplug operations. Developed RL-based control policies and integrated computer vision for connector detection and pose estimation.

NVIDIA Isaac Sim 4.5 Reinforcement Learning Computer Vision IIWA 14 Robot

Projects

Task-Aware Local RAG for Robot Procedure and Code Generation

Built a local RAG system using Ollama + FAISS to extract grounded contingency procedures & generate robot code with 40% reduction in hallucinations.

Ollama FAISS RAG Local LLM

Achievements:

  • Recall@5: 84% and MRR: 0.78 on 20+ documents
  • Deployed for on-floor fault recovery support
  • Generated control code for recovery actions with low latency

Real-Time Stock Market Ticker - Full Stack Web & Android App

Developed a real-time stock market ticker and portfolio management system with seamless UI/UX.

MERN Stack React.js Node.js MongoDB Android
View Code Live Demo

CiteLlama – Improving Citations with Context

Fine-tuned Llama 3.1 8B using LoRA on academic citations to generate accurate, grounded citations with 11% boost in quality.

Llama 3.1 LoRA Fine-tuning Academic Citations

Features:

  • Implemented citation formatting for CVPR, AAAI, IEEE, ACM
  • 11% boost in citation quality (BLEU + relevance)
  • 50K+ academic citations dataset

Dexterous Manipulation - Throwing & Catching

Developed an RL-based framework for dexterous manipulation tasks using MuJoCo environments with hand models.

Reinforcement Learning MuJoCo SAC Dexterous Manipulation

Technologies:

  • Leap Hand & Wonik Allegro models
  • Soft Actor-Critic (SAC) algorithm
  • GPU-accelerated training

MRI Data Pipeline & GAN-based Augmentation

Engineered MRI data pipeline for 10,000+ scans with GAN-based augmentation for anomaly detection.

GAN Medical Imaging Data Pipeline MRI

Results:

  • Improved MRI image quality (1.5T → 3T equivalent)
  • 10% improvement in classification accuracy
  • Low-latency processing pipeline

Vision-Based Tool Wear Detection System

Developed computer vision system for real-time tool wear detection in manufacturing environment.

Computer Vision OpenCV Manufacturing Quality Control

Features:

  • Real-time processing on production line
  • Integrated with OPC UA framework
  • Reduced tool replacement downtime