Mehdi Sadeghian

Mehdi Sadeghian

Autonomous Systems & Deep Reinforcement Learning Researcher

I am a graduate researcher with a strong background in Mechatronics and Control Theory. My research focuses on developing intelligent, safety-aware decision-making algorithms for autonomous vehicle navigation, multi-agent trajectory forecasting, and deep generative computer vision in dynamic urban environments.

Simscape & Robotics

SCARA Manipulator: Dynamics & Simscape Control

SCARA CAD integration, analytical inverse kinematics, trajectory generation, and independent-joint PID tracking in MATLAB/Simulink Simscape Multibody.

Simulink Simscape MATLAB Joint PID SolidWorks

Pedestrian Trajectory Prediction with Social-STGCNN

ETH / UCY Benchmark

Pedestrian Trajectory Prediction (ST-GCNN)

Multi-agent trajectory forecasting using Spatio-Temporal Graph Convolutions and CVAE-GMM multimodal latent space sampling for anticipating diverse human navigation choices in autonomous driving.

PyTorch ST-GCNN CVAE-GMM ETH / UCY

Deep RL

DQN for Dynamic Prompt Modifier Selection

A DQN agent that selects discrete textual prompt modifiers for a frozen TinyLlama-1.1B model on BoolQ, balancing answer correctness and response length.

PyTorch DQN TinyLlama HuggingFace BoolQ

YOLOv11 & GenAI

Helmet Detection & Video Generation

Two-class YOLOv11 helmet detection paired with text-to-video generation and framewise inference on four short synthetic rider clips.

YOLOv11 PyTorch Diffusion mAP50: 83%

Function Approximation and Deep Q-Networks in RL

Deep RL

Function Approximation & Deep Q-Networks

Bridging classical tabular methods to continuous state spaces via Linear Semi-Gradient TD(0), Tile Coding, and Deep Q-Networks (DQN) with experience replay and target networks.

PyTorch DQN Tile Coding Semi-Gradient TD

Unpaired Image-to-Image Translation with CycleGAN

Generative Vision

Unpaired Image Translation with CycleGAN

Deep generative image-to-image translation pipeline implementing dual PatchGAN discriminators with ResNet-based cycle-consistency and identity loss constraints on Horse2Zebra.

PyTorch CycleGAN PatchGAN ResNet Generator

Dyna and Dyna-Q planning in reinforcement learning

Model-Based RL

Model-Based Planning with Dyna & Dyna-Q

Direct Q-learning updates combined with simulated model-based planning in a deterministic 5×5 Grid World.

Dyna-Q Dyna-Q Model-Based RL Planning

Medical AI

Cancer Histopathology Classification

Comparative five-class histopathology image classification study benchmarking Vision Transformers (ViT-B/16) against ResNet-50 with transfer learning.

Vision Transformer ResNet-50 LC25000 Transfer Learning

Dynamic Programming

Jack's Car Rental: Dynamic Programming & MDPs

Exact Markov Decision Process resolution via Policy Iteration, Policy Evaluation, and Value Iteration for Poisson-distributed resource allocation and inventory management.

Policy Iteration Value Iteration MDP Poisson Dist.

GridWorld Exploration Policies Benchmark

Exploration Benchmark

GridWorld Exploration Dynamics & Policies

Comprehensive empirical evaluation of ε-Greedy, Upper Confidence Bound (UCB), and Stochastic Gradient exploration strategies on custom obstacle GridWorld MDPs.

UCB Gradient Bandit Epsilon-Greedy Heatmaps

Core RL

Tabular RL Methods Suite (SARSA / Q-Learning)

GridWorld comparison of first-visit Monte Carlo and TD(0) prediction with SARSA and Q-Learning control.

Q-Learning SARSA TD(0) Monte Carlo Prediction

Bandit Algorithms

Non-Stationary Multi-Armed Bandits

Comparative study analyzing constant step-size parameter (exponential recency-weighted average) versus standard sample-average tracking in non-stationary reward distributions.

Multi-Armed Bandit Step-Size α Sample-Average

Edge Vision

Face Mask Classification (MaskCNN)

Lightweight custom CNN for binary classification of masked and unmasked face images, with confusion-matrix and ROC analysis.

Custom CNN PyTorch Binary Classification ROC Analysis

NLP & Text

Amazon Fashion Sentiment Classification

Bag-of-Words sentiment experiment comparing logistic regression, linear SVM, and a decision tree on a balanced single-product review sample.

Scikit-Learn Bag-of-Words Sentiment Analysis

Health Analytics

U.S. Medical Cost & Insurance Risk Modeling

Exploratory insurance-charge analysis comparing region-only linear regression with a smoker, age, and BMI linear model.

Linear Regression EDA Scikit-Learn

Financial AI

Credit Card Fraud EDA & Risk Profiling

Exploratory analysis of one million synthetic transaction rows, covering class balance, correlations, authentication indicators, and online/offline fraud rates.

Pandas EDA Correlation

Research & Academic Inquiries

I am always open to discussing novel research directions, exchanging ideas on Autonomous Systems & Deep Learning, and actively exploring prospective Ph.D. positions and academic collaboration opportunities.

Academic Background

M.Sc. in Mechatronics Engineering

B.Sc. in Electrical Engineering (Control Theory)

Research Areas & Expertise

Safe Autonomous Navigation & Control

Multi-Agent Trajectory Forecasting

Computer Vision & Generative AI

Medical Image Analysis & ViTs

Featured Research & Engineering Repositories

SCARA Manipulator: Dynamics & Simscape Control

Pedestrian Trajectory Prediction (ST-GCNN)

DQN for Dynamic Prompt Modifier Selection

Helmet Detection & Video Generation

Function Approximation & Deep Q-Networks

Unpaired Image Translation with CycleGAN

Model-Based Planning with Dyna & Dyna-Q

Cancer Histopathology Classification

Jack's Car Rental: Dynamic Programming & MDPs

GridWorld Exploration Dynamics & Policies

Tabular RL Methods Suite (SARSA / Q-Learning)

Non-Stationary Multi-Armed Bandits

Face Mask Classification (MaskCNN)

Amazon Fashion Sentiment Classification

U.S. Medical Cost & Insurance Risk Modeling

Credit Card Fraud EDA & Risk Profiling

Technical Skills & Frameworks

Deep Learning & AI

Robotics & Control

Engineering & Deployment

Languages & Mathematics

Teaching & Industry Experience

Teaching Assistant

Teaching Assistant

Machine Learning Engineer Intern

Research & Academic Inquiries