I'm Hemanth Patel from IIT BHU Varanasi. Iām passionate about Human-Robot Interaction, building intelligent robotic software, and AI-applied robotics. I recently completed my B.Tech, and have worked with Drobot.Inc for the past two years as a Robotics Software Engineering Intern.
š Download Resume
Physics-Informed Neural Networks (PINN)-based Obstacle Avoidance
Rudrashis Majumder, Hemanth Patel, Sri Siddarth Chakaravarthy P, Samahith S A, Suresh Sundaram.
"OA-PINN: Efficient Obstacle Avoidance for Autonomous Vehicle Safety With Physics-Informed Neural Networks,"
Accepted and presented at IEEE CONECCT 2024.
Read Paper
B.Tech Student ā IIT BHU Varanasi
Research Intern ā Artificial Intelligence and Robotics Laboratory, IISc Bangalore (Prof. Dr. Suresh Sundaram)
Robotics Software Engineer ā Drobot.Inc
Secretary, Robotics Club ā Science and Technology Council, IIT BHU
Compared LSTM, GRU, and CNN on Mel spectrogram and MFCC features, achieving up to 95% accuracy. Used Librosa preprocessing and real-time speech enhancement. Mel spectrogram data outperformed wav2vec2 on meta-audio features.
Demo Video
Winner of AI Olympics International (Simulation stage). Implemented SAC with MLP policy and custom energy-based rewards for effective swing-up and stabilization within 20 seconds in OpenAI Gym.
GitHub | Demo Video
Developed a model combining TrOCR for visual feature extraction and T5-small for entity prediction on 250K images. Utilized Hugging Face Transformers for fine-tuning, achieving an F1 score of 0.68 through iterative training.
GitHub
Developed a robotic arm with 5 cm pick-and-place precision using MoveIt, ROS, and YOLOv8 for 360Ā° inspection.
Implemented voice command control with 90% accuracy. B.Tech project supervised by Dr. Amit Tyagi, IIT (BHU) Mechanical Department.
GitHub | Report | Video presentation
3D-printed quadruped designed for seamless human-robot interaction with fast response software.
Implemented purely in Python without ROS, focusing on simplicity and real-time control.
Demo Video
Led a team to integrate Neural-Fly model using Domain Adversarially Invariant Meta-Learning for wind-aware control with minimal data.
Implemented MPC to handle actuation limits, enabling precise trajectory tracking and robust performance in varying wind conditions. Supervised by Dr. Shyam Kamal, IIT (BHU) Electrical Dept.
GitHub | ReportEmail: hemanth.patel.mec21@iitbhu.ac.in
Phone: +91 9010765370
LinkedIn: linkedin.com/in/hemanth-patel-721361235