Seyed Ahmad Hosseini

Seyed Ahmad Hosseini

AI & Robotics Researcher • Control Engineering

Amirkabir University of Technology
Tehran, Iran

About

Control Engineer with advanced expertise in Artificial Intelligence and Machine Learning, specializing in autonomous systems, robotics, and computer vision. B.Sc. student in Electrical Engineering (Control) at Amirkabir University of Technology, focusing on Reinforcement Learning and intelligent control systems.

Publications

BLIP‑FusePPO: Vision‑Language Model Enhanced Multimodal Reinforcement Learning for Autonomous Lane‑Keeping

First Author IEEE Transactions on Intelligent Vehicles Under Review
  • Developed a multimodal RL framework integrating semantic embeddings from BLIP with geometric states, LiDAR data, and control feedback for autonomous driving.
  • Designed a lightweight state representation that retains semantic awareness while removing runtime VLM inference, enabling real‑time deployment.
  • Achieved a 54.5% RMSE reduction over DDPG and 44.4% improvement over VL‑SAFE in varied driving scenarios.
  • Conducted extensive simulations validating policy stability and adaptability.

Ongoing Research

IRL-DAL: Hybrid RL with a Safety-Aware Diffusion Planner for Autonomous Driving

Lead Researcher Jan 2025 – Present
  • Architected a novel hybrid framework (IRL-DAL) by synergizing IL, IRL, and PPO, achieving a 50% reward increase over baselines.
  • Engineered a safety-aware Diffusion Planner as a fallback, generating feasible actions to significantly reduce collision rates.
  • Designed a Learnable Adaptive Mask (LAM) for dynamic visual attention, leading to a precise lateral deviation of only 0.105m.
  • Implemented a two-phase curriculum (Imitation → Mixed RL) with prioritized replay, ensuring stable policy convergence in Webots.

Selected Projects

Multimodal RL for Autonomous Driving in Webots

Reinforcement Learning – Autonomous Driving
  • Designed a deep RL framework integrating BLIP vision–language embeddings with LiDAR, control feedback, and geometric states, enabling robust lane‑keeping in Webots simulations.
  • Optimized state representation to maintain semantic awareness while removing runtime VLM inference overhead, achieving real‑time performance.

A* Path Planning for Obstacle Avoidance with Computer Vision in Webots

Computer Vision – Path Planning
  • Developed an A*-based path planning system for real-time obstacle avoidance in Webots simulations.
  • Applied computer vision for dynamic environment perception, improving autonomous navigation decisions.

End‑to‑End Generative AI Systems for Multilingual Alignment (RLHF)

Large Language Models – Multilingual NLP
  • Built an end-to-end RLHF pipeline with FLAN‑T5, leveraging PPO, DPO, and GRPO to improve coherence and human-preference alignment.
  • Fine-tuned English–Persian translation models using LoRA and QLoRA, achieving competitive BLEU/ROUGE scores while minimizing computational cost.

Multimodal Speech Processing with Wav2Vec2

Deep Learning – Speech Processing
  • Developed a multimodal pipeline for 8-class Speech Emotion Recognition and ASR, processing raw audio and corresponding text transcriptions.
  • Fine-tuned Wav2Vec2 models (facebook/wav2vec2-large-xlsr-53 for SER, facebook/wav2vec2-large-960h for ASR) on CREMA-D, RAVDESS, TESS, and SAVE datasets with enhanced preprocessing, augmentation, and class mapping.

Transformer for Twitter Emotion Detection

Deep Learning – NLP
  • Designed a Transformer in PyTorch for multi-class tweet emotion detection, leveraging GloVe embeddings and customized preprocessing.
  • Integrated trainable positional encoding, oversampling for class balance, and dynamic learning-rate scheduling, enhancing model accuracy and generalization.

Self-Balancing Two‑Wheeled Robot

Robotics – Autonomous Systems
  • Built a modular two-wheeled self-balancing robot with Arduino UNO and L298N motor driver, using dual 25GA 330RPM DC motors and MPU6050 IMU for real-time tilt correction and stable motion under disturbances.

Autonomous Navigation Algorithms in Webots

Robotics – Autonomous Systems
  • Developed a proportional controller for line-following robots in Webots, diagnosing control limitations and proposing PID with intersection detection to improve stability and accuracy.
  • Built a maze-solving robot using iterative DFS in Webots and validated successful navigation; proposed BFS for shortest-path optimization.

RISC‑V Single‑Cycle Processor

Computer Architecture – Digital Design
  • Implemented a modular single‑cycle RISC‑V CPU in VHDL supporting R‑type (ADD, SUB, AND, OR) and I‑type (ADDI, ANDI, ORI, LW, SW) instructions, including PC, memory units, register file, ALU, immediate generator, and control unit.
  • Validated CPU functionality via comprehensive testbenches across arithmetic, logical, immediate, and memory operations, ensuring hazard-free execution and correct control signaling.

Teaching & Workshop Experience

Intensive MATLAB Programming Workshop

Instructor & Organizer Jun 2023
  • Designed and taught an intensive MATLAB workshop focused on control system implementation.
  • Led hands-on exercises and interactive problem-solving sessions for undergraduate engineering students.

Computer Vision

TA (Supporting) Jan–Jul 2025
  • Assisted in coursework and student support.

Electronics II

Lead TA Sep–Dec 2024
  • Led problem‑solving sessions during lectures and supported grading.
  • Assisted students with circuit design and analysis.

Linear Control Systems

TA (Supporting) Sep–Dec 2024
  • Supported tutorials and grading.
  • Provided guidance to students in applying control theory concepts.

Instrumentation

TA (Supporting) Jan–Jul 2025
  • Assisted in coursework and student support.

Artificial Intelligence and Digital Transformation

TA (Supporting) Sep–Dec 2025
  • Assisted in coursework and student support focusing on AI applications in industry.

Modern Control Systems

TA (Supporting) Sep–Dec 2025
  • Assisted in coursework and provided support for advanced control theory classes.

Industrial Control

TA (Supporting) Sep–Dec 2025
  • Assisted in coursework and student support for industrial automation topics.

Honors & Awards

  • 2021: Ranked 270th, Top 0.8% among 35,000+ Region 3 participants in the Iranian University Entrance Exam (Konkoor) for B.Sc. in Engineering
  • 2018: Admitted to Abu Taleb Bagheri High School, Member of NODET (National Organization for Development of Exceptional Talents)

Certifications

Image Processing and Computer Vision with OpenCV

Maktabkhooneh 2024 Online

Mastering Raspberry Pi Pico Programming with MicroPython

FaraDars 2024 Online

Mastering Git, GitHub, and GitLab: Your Ultimate Guide to Version Control

FaraDars 2024 Online