Curriculum Vitae

Yaozeng Huang · 黄耀增13325905201@163.comXinjiang University · Computer Sciencegithub.com/hyyyyyyz

TIP

This page mirrors my latest CV. A PDF copy is available on request via email.

Education

Xinjiang University · B.Eng. in Computer Science & Technology (211 Project, Double-First-Class discipline)    2023.08 – 2027.07

• GPA: 3.9 / 5.0; class rank: 17 / 120 (top 14.2 %)
• Languages: CET-6 444
• Internal member of the Xbotics embodied-AI open-source community

Research Experience

1. Force-Aware VLA Models for Dexterous Manipulation · Tsinghua SIGS Oct 2025 – Present

1. Multi-modal real-robot platform. Built a bimanual platform with Songling 7-DoF arms + grippers and a Jetson AGX Orin compute box, equipped with two D405 wrist cameras and a head-mounted D435; collects joint position / torque proprioception and depth streams for contact-rich tasks such as optical-module insertion and block placement.
2. Reproducing torque-aware VLA models. Reproduced and extended pi0.5 and TA-VLA, with a focus on injecting force/torque modalities into base vision-language-action models. Surveyed and contrasted diffusion-policy, flow-matching, autoregressive, and hybrid VLA architectures along the axes of high-frequency action generation and contact-state modelling.
3. In progress. Investigating the geometric failure modes of VLA policies and an RGB-D privileged-geometry distillation method that uses action-conditioned 2.5D interaction maps to repair contact / placement failures under occlusion and depth ambiguity. Targeting a first-author submission.

2. RL for Dynamic-Obstacle Local Navigation · Eastern Institute of Technology, Ningbo Jan 2026 – Present

1. Problem extension. Extending the DRL-DCLP local-planning line of work from static to dynamic obstacle settings; reformulating the avoidance MDP for robots of varying footprint, with matching simulation environments, state spaces, and training tasks.
2. Curriculum design. Built an 8-stage curriculum on StageRos: each successive stage scales the map by ~80%, increases static density, and introduces a hyperparameterised number of dynamic obstacles.
3. Temporal dynamic encoding. Implementing a temporal dynamic encoding module for multi-obstacle state and fusing it into both policy and value networks to improve generalisation in cluttered, time-varying environments.
4. Goal. Targeting a co-authored submission to T-RO / TMECH.

3. RL Locomotion & LiDAR Navigation for Quadrupeds · XJU Innovation Lab Jun 2025 – Present

1. RL training. Trained PPO and N-P3O policies in Isaac Gym / Isaac Lab for multi-terrain quadruped locomotion, generalising across ≥ 3 terrain types.
2. Sim-to-Real. Built and tuned a self-designed quadruped; used HIMLoco for MuJoCo sim-to-sim and finally sim-to-real, sustaining > 10 N pushes with stable gait control — walking 0.8 m/s, jogging 2 m/s, crawling — across 4 indoor / outdoor terrain classes and slopes.
3. LiDAR navigation. SLAM with fast_lio and point_lio for indoor / outdoor scenes; ROS 2 Nav2 stack with A* / Dijkstra global planning and AMCL relocalisation for downstream tasks.
4. Outcome. ICIC 2026 (CCF-C, first author) — accepted, Oral Presentation.

4. Deep-Learning Strawberry Disease & Pest Monitoring · Xinjiang University Jun 2024 – Jan 2026

1. Detection. YOLOv8-based detection pipeline with attention modules and an improved feature-fusion neck — +15 % recall on small targets and +25 % precision; pruned and quantised for embedded deployment.
2. Manipulation. Inverse-kinematics solver for a tracked-rover-mounted arm; coordinated chassis-arm motion to acquire clear, comprehensive plant imagery from preset observation viewpoints.
3. Outcome. National Innovation & Entrepreneurship grant — Outstanding close-out · 1 design patent (1st author) · 1 software copyright (3rd author).

Honors & Awards

Date	Award	Level
2025.07	24th National Robotics Competition (ROBOCON) — Bionic Quadruped Obstacle Race	National First Prize
2025.07	24th National Robotics Competition (ROBOCON) — Bionic Quadruped Off-road Race	National First Prize
2025.08	2025 National Undergraduate Electronic Design Contest (NEDC)	National Second Prize
—	Total: 9 national-level + 3 provincial-level competition awards	—

Skills

• Languages: Python · C++ · MATLAB · Java
• Operating systems: Linux · ROS 2
• Simulation: Isaac Gym · Isaac Lab · MuJoCo · Gazebo
• Robot learning frameworks: LeggedGym · OpenVLA · LeRobot · PyTorch
• Tooling: Conda · Git · Docker
• Familiar with: common RL algorithms (PPO / N-P3O / SAC, …) and mainstream VLA paradigms

Open Source & Community

• Personal repositories: https://github.com/hyyyyyyz
• Internal member of the Xbotics embodied-AI open-source community