Projects | Zunzhe Zhang

QDGS: 3D Scene Reconstruction with High-Dimensional Surface Primitives

Sun, 01 Dec 2024 00:00:00 +0000

Extended 2D Gaussian Splatting by replacing planar primitives with high-dimensional curved surface primitives.

Designed geometric routines for ray-surface intersection, differentiation, and bounding box computation. The approach improved reconstruction error from 0.78 to 0.72 while using fewer primitives, showing better geometric representation efficiency.

CyberDog 2 Dribbling with Deep Reinforcement Learning

Sat, 01 Jun 2024 00:00:00 +0000

Trained quadrupedal and bipedal dribbling policies for CyberDog 2 using reinforcement learning.

Implemented a vision-based goalkeeper module and integrated a complete 2v2 match environment. The system achieved 2nd place out of 16 teams in a university robotics competition.

LLM-Enhanced RPG Game with Context-Aware NPC Agents

Sat, 01 Jun 2024 00:00:00 +0000

Built an LLM-driven RPG prototype using GPT-4o and RPG Maker to enable context-aware NPC dialogue generation.

Designed prompt templates and game-state interfaces that allowed NPCs to condition responses on player progress, dialogue history, and task context. The project explored adaptive response generation and interactive narrative control.