Zhiyao He

Profile

I am an AI engineer working on embodied AI systems. I build robot pipelines that connect perception, language models, planning and control, from simulation (Isaac Gym, MuJoCo, Gazebo) to real hardware. My recent work is a voice-controlled navigation and manipulation system on the Unitree Go2-W quadruped robot. I am seeking a PhD position in robot learning, especially on how robots can learn manipulation skills that work reliably in the real world.

Education

Grade: 82.33 / 100 (research-based degree)

• Thesis work on machine learning for runoff forecasting; published two peer-reviewed journal papers and obtained one Chinese invention patent.
• Relevant courses: Information Acquisition and Processing, Optimization. Certificate: Deep Learning Specialization (DeepLearning.AI).

CGPA: 3.65 / 4.00

• Finished the four-year program in three years.
• Relevant courses: Control Theory, Circuit Design, C Programming, Digital Signal Processing.

CGPA: 4.00 / 4.00

• President's List for 12 terms.
• Relevant courses: Python, Java, Databases, Operating Systems, Software Engineering, Computer Networks.

Publications and Patent

• Zhiyao He, Mingxiang Yang, Hao Wang, et al. “Annual Average Runoff Ensemble Forecast for Jinping I-Stage Hydropower Station Based on Elman Neural Network.” Water Resources and Power, 2017.
• Chenming Li, Lei Zhu, Zhiyao He, et al. “Runoff Prediction Method Based on Adaptive Elman Neural Network.” Water, 2019.
• Zhiyao He (Patent holder). “Medium- and Long-term Runoff Forecasting Method Based on Mutual Information, Kernel PCA, and Elman Neural Network.” Chinese National Invention Patent, CN107463993B, granted 2020.

Work Experience

• Built the perception, planning and control stack for voice- and text-controlled navigation and manipulation on the Unitree Go2-W quadruped robot with a D1 arm (details in Projects).
• Implemented the closed-loop control system of the AKI humanoid robot: a reinforcement learning policy layer, a joint target layer, and PD motor control.
• Designed the AKI robot's power management system, covering power input, distribution and conversion, and the load execution layer.
• Integrated YOLO-based perception with ASR, LLM and TTS modules for face recognition and voice interaction.
• Set up simulation environments and ran locomotion training for the AKI and Unitree G1/R1 humanoid robots in Gazebo, MuJoCo, and Isaac Gym.

• After Huawei, I planned to return to research. During this period my second journal paper (Water, 2019) was published and my invention patent was granted (2020).
• From 2021 to 2024, I completed a full computer science degree at University of the People (GPA 4.00/4.00, see Education) to retrain from signal processing toward AI and robotics. Part of this period overlapped with the COVID-19 lockdowns in China.
• Admitted to the M.Sc. in Data Science at Trinity College Dublin (2025 intake), but did not enroll.
• After that, I prepared for AI engineering interviews and joined WellWin Technology in March 2026.

• Worked on computer vision systems for face and vehicle recognition, mainly on development and performance optimization.
• Debugged and tuned the systems together with other teams.

• Built a complete runoff prediction pipeline: data collection and cleaning, feature engineering, model development, hyperparameter tuning, training and evaluation.
• This work produced two journal papers and one invention patent (listed above).

Selected Projects

• Perception: YOLO11 object detection with a wrist-mounted depth camera, plus 2D-to-3D coordinate transformation to locate the target in the world frame.
• Navigation: SLAM-based autonomous navigation to the target position.
• Language: a task planning pipeline with ASR, LLM, OpenClaw and TTS, so the robot can be controlled by natural language and reply by voice.
• Manipulation: a grasping pipeline based on forward/inverse kinematics and Jacobian-based motion control.
• Developed in ROS 2, tested in MuJoCo simulation and on the real robot.

• Real-time control loop for locomotion and interaction: an RL policy generates joint positions, a joint target layer does scaling, integration and clipping, and a PD layer drives the motors.
• ASR–LLM–TTS pipeline for natural voice interaction.

• An in-store retail assistant: customers ask about products, plans, and pricing by voice.
• Built a fully offline pipeline on a Jetson for events with unreliable networks: a bridge node streams the X2's microphone audio to a local FastAPI backend, which runs local ASR (FunASR/Paraformer-zh), a local LLM (Qwen2.5-7B via Ollama), pgvector RAG over the product database (PostgreSQL), and local TTS (Sherpa-ONNX / Piper).

Technical Skills

• Programming & SystemsPython, C++, Linux, Docker, Git
• RoboticsROS/ROS 2, Isaac Gym, Gazebo, MuJoCo, SLAM, Navigation, Localization, Perception
• AI & Robot LearningPyTorch, Machine Learning, Computer Vision, Reinforcement Learning, ASR, TTS, LLMs
• Robot PlatformsUnitree G1, Unitree Go2/Go2-W, AgiBot Lingxi X2

Research Interests

• Robot Learning and Reinforcement Learning
• Vision-Language-Action Models
• Robot Manipulation and Grasping
• Robotics Simulation, SLAM and Navigation

Honors and Awards

• President's List, University of the People (12 terms)2021–2024
• Outstanding Contribution Award, Huawei Nanjing R&D Institute2019
• Outstanding Graduate Student, Hohai University2017
• Hohai University Scholarship2015–2017
• Anhui Agricultural University Scholarship2012

Teaching and Outreach

• Designed a humanoid robotics curriculum for Macao secondary school students, progressing from no-code tools (LingXin, LinkCraft) to ROS 2 programming with the AimDK SDK, and finally to multi-platform integration.

• Designed the task and rules for a humanoid robot ball-shooting competition, testing vision recognition, trajectory planning and 3D printing.

• Answered prospective students' questions about admissions and academics, helping applicants with limited educational opportunities gain access to quality higher education.