Prabin Kumar Rath

I earned my Master's in Robotics and AI from Arizona State University in Spring 2024 and currently serve as an MLOps Engineer at Experian. My core interests include robot learning, 3D perception and software engineering.

For my Master's thesis, I worked on motion planning for robotic manipulators at Logos Lab advised by Prof. Nakul Gopalan. I developed deep learning-based behavior cloning algorithms that zero-shot generalize to configuration-space of unseen robotic arms. My work involved generating large-scale synthetic datasets, training robot policies, and developing 3D collision detection models for neural motion planning. I also collaborated with Prof. Hongbin Yu at the EVSTS Lab, focusing on the application of 3D object detection and tracking algorithms for quantifying AI safety. Specifically, I analyzed unsafe situations for vulnerable road users such as pedestrians and cyclists at urban traffic intersections.

I received my Bachelor's in Computer Science and Engineering from National Institute of Technology, Rourkela in Spring 2020. During my undergrad, I interned at the University of Calgary with Prof. Alex Ramirez Serrano, to work on 3D moving object detection techniques for collision-free robot navigation in indoor environments. Prior to that, I interned with Prof. Dilip Kumar Pratihar at IIT Kharagpur, where I developed PID controllers for teleoperation of a custom bipedal robot to mimic human gait patterns. After graduation, I worked for two years at Wells Fargo as a software engineer, developing scalable OCR pipelines for data extraction from scanned forms and handwritten documents.

For my next career milestone, I am actively seeking a PhD position in robot learning, with plans to begin in Fall 2025.

CV  |  Github  |  Scholar  |  LinkedIn  |  RG

A novel neural policy that solves motion planning problems zero-shot for unseen robotic manipulators. We demonstrate for the first time that configuration-space behavior cloning policies can be learned without embodiment bias and that these learned behaviors can be transferred to novel unseen embodiments in a zero-shot manner.

Real-time VRU safety metrics evaluation strategies for urban traffic intersections.

Leader-follower pair identification and metrics evaluation from aerial drone videos.

Complex-YOLO 3D detection and EKF tracking pipeline for measuring intersection safety metrics with CAROM dataset as pseudo ground truth.

An IoT health kit for hospitals to assist doctors with medical diagnosis.

Detection and Tracking of Moving Objects (DATMO) algorithm for robot navigation in dynamic environments.

paper | code

Room Automation Module for switching electrical appliances remotely through Web, Bluetooth, and IR.

A robot for playing the game of chess physically with an user. Powered with strong chess engines and interactive UI it provides all virtual game features while ensuring the authenticity of the original board game.

Multi-robot swarm controller for pattern formation using Signed Distance Field (SDF) of hand drawn images.

Implementation of T-REX and D-REX Inverse Reinforcement Learning (IRL) algorithms for learning form suboptimal demonstrations.

Deep Reinforcement Learning with PyTorch and OpenAI-Gym. Implementation of Deep-Q-Learning and Dueling Double Deep-Q-Learning Algorithms.

An AI agent that utilizes Monte Carlo Tree Search (MCTS) for exploration and exploitation but switches to customized reflex actions at critical zones.

Implementation of ST-ResNet model for predicting traffic flow on BikeNYC and TaxiBJ datasets.

A tutorial for integrating ROS navigation and mapping stack with your custom differential drive robot.

Trajectory optimization for the cart pole swing-up problem using Direct collocation, and balancing the inverted pendulum on the cart using LQR control.

Simple LSTM model for classifying colloquial english sentences onto motion commands. Curated text to command dataset for model training using Tensorflow.

Mathematical modelling of end effector pitch. Closed-loop PID control for canceling the base motion effects on the end effector.

Customized replica of Turtlebot arm for pick and place tasks. ROS Control and MoveIt stack. Gym compatible environment for training RL agent in Gazebo.

6DOF holonomic robot with integrated vision sensors for carrying out planned missions autonomously in underwater environments.

A 2D Dot matrix printer is a DIY project made from old CD drives.

A Bluetooth controlled robot with on board FSR (Force Sensitive Resistor) for weight detection. It has indication LEDs to prompt the user about the detected weights.

A 8*8*8 LED cube made using 9 shift registers 74HC595 and Arduino Nano.