Projects

MLP, CNN, RNN, GNN, VAE, GAN, Diffusion — Build a complete deep learning knowledge system from scratch

MDP, DP, MC, TD, DQN, Policy Gradient, PPO, SAC — Systematically master core RL theory

Value learning, exploration, evaluation, generalization, world models, search & games — A complete Game AI course through Atari

From Go rules to hand-coding AlphaGo / AlphaGo Zero — game-tree search → MCTS → policy/value networks → SL → RL self-play

CFR → Libratus → Pluribus → StarCraft MARL — A unique direction on imperfect information games

SLAM, path planning, coverage algorithms, RL navigation, perception — Make a vacuum robot clean on its own

Dynamics, RL grasping, vision picking, ACT imitation learning, LLM chess — From kinematics to intelligent manipulation

Isaac Lab, PPO, quadruped→humanoid, RMA, CPG, AMP, distillation — Make a humanoid robot stand up and walk

Flight dynamics → PID → path planning → VIO → SLAM → RL flight policy → formation → Sim2Real — Autonomous quadrotor racing through gates

Game theory → IQL → CTDE (QMIX/VDN) → MAPPO/MADDPG → communication learning → self-play → hide-and-seek → swarm formation — Filling the MARL gap

Core AI security topics from adversarial examples to LLM safety, covering attacks, defenses, privacy, fairness, and trustworthy AI

Complete learning path from attention mechanisms to multimodal LLMs, covering pretraining, fine-tuning, RAG, agents, and more

Complete path from diffusion models to video generation, covering Stable Diffusion, ControlNet, LoRA, FLUX, AnimateDiff and more

Complete autonomous driving pipeline from sensors to end-to-end driving in the CARLA simulator

Full RLHF/DPO pipeline from reward modeling to alignment evaluation, balancing theory and practice

Aligned with CMU 16-831, covering Inverse RL, Safe RL, Meta-RL and advanced robot learning methodologies

Systematic methodology for deploying simulation-trained policies to real robots

MAPF / CBS / task allocation / auction mechanisms / fleet management / KPI evaluation — From single-robot nav and MARL to large-scale warehouse orchestration

Drive robot perception, planning and manipulation with LLM/VLM, from SayCan to VoxPoser

From classical NPC AI to modern universal agents — SIMA, Game-TARS, LLM-driven NPCs, PCG, the game AI capstone

VAE/GAN → diffusion → DreamerV3/RSSM → Diffusion Policy → video world models — Agents that learn in imagination

Cognition × Perception × Action × Learning × Deployment — Build a complete robot system that understands, perceives, plans, acts, and learns

Shadow/Allegro Hand, grasp planning, RL grasping, in-hand manipulation, TACTO tactile, dual-hand coordination — Natural sequel to 321

ERA5 / Sentinel-2 / NetCDF data fundamentals + satellite CNN classification + short-term weather ML forecasting + disaster monitoring — AI for the planet beneath our feet

Astronomy data fundamentals + galaxy morphology classification (Galaxy Zoo) + exoplanet detection (Kepler) + gravitational wave detection (LIGO) — AI for the cosmos

From financial data processing to multi-strategy portfolios, covering factor research, ML signal extraction, backtesting, and portfolio optimization for AI-native quant research.