Zhenqi Wu

Ph.D. Student · Robotics & Multimodal AI

Tampa, FL | zhenqi@usf.edu

About Me

Zhenqi Wu

I am a Ph.D. student in Electrical Engineering at the University of South Florida. I study how robots use vision and language to decide where to go and how to get there safely. My research in multimodal and embodied AI combines vision-language models (VLMs), vision-language-action (VLA) policies, and motion planning. I test these methods in simulation and on a BlueROV2 underwater robot, with a focus on autonomous navigation and exploration.

Education

  • Aug 2025 – Present: Ph.D. in Electrical Engineering, University of South Florida · Expected May 2029
  • Aug 2023 – May 2025: Doctoral studies in Computer Science, Arizona State University
  • Aug 2021 – May 2023: M.S. in Electrical & Computer Engineering, University of Florida
  • Sep 2016 – Jul 2020: B.S. in Electrical Engineering & Automation, Dalian Maritime University

Research & Projects

DREAM · Underwater habitat monitoring

I developed a VLM-guided exploration system that uses images, depth, and a map of the seafloor to decide where to search next. In physical BlueROV2 trials, DREAM located every object of interest with zero collisions.

CORAL · Hierarchical planning for underwater exploration

CORAL uses a VLM to select waypoints and a dynamics-aware local planner to reach them. Across 10 simulated environments, it achieved 94.28% coverage with zero collisions. In physical BlueROV2 trials, CORAL found all objects of interest with zero collisions.

Project website

AQUAACT · Underwater robot learning (ongoing)

I am currently training VLA policies on language-annotated teleoperation demonstrations from a physical underwater robot and a simulator. The dataset pairs video and vehicle sensors with human joystick commands. Closed-loop policy evaluation is planned.

APPLV · Learning to tune a navigation planner

We fine-tuned a vision-language model to predict parameters for classical motion planners. We evaluated the approach in simulated BARN environments and on a physical Clearpath Jackal.

CAVE-NAV · 3D cave navigation

We combined RGB images, depth maps, and sonar clearance measurements to infer a safe direction of travel through confined underwater passages. We evaluated the system in simulated caves.

Earlier projects

GuardianML – Recommender System for Privacy-Preserving ML

Co-developed an open-source system that automatically recommends optimal configurations for privacy-preserving machine learning (PPML) deployments. Built structured datasets and conducted algorithm comparisons.

Additional Projects

  • Bio-inspired STDP Neural Networks for Real-Time Object Detection
  • Traffic Statistics via DeepSort & YOLOv5 for Control System Integration
  • Vehicle Safety Alert using DeepLabv3+ and Camera Fusion
  • ESO-Based Adaptive Control for Unmanned Surface Vehicles

Selected Publications

Tech Stack

Python C++ ROS PyTorch Isaac Sim MuJoCo OpenCV

Connect with Me