CS undergrad @ CUMT (Rank 9/185) · Research Intern @ Tsinghua SIGS (Prof. Zhi Wang's group)
Working on Spatial Intelligence & Embodied AI — long-horizon robot manipulation, VLA state memory, and spatial reasoning.
- 🔭 Now: training-free VLM planning & long-horizon VLA control (RLBench · LIBERO · SIMPLER)
- 🦾 Hardware: SO-101 arm (teleop + π0 fine-tuning), Unitree G1 / Go2 locomotion in MuJoCo
- 🌱 Exploring: bridging VLM reasoning and low-level robot control
- 🏆 Honors: "Challenge Cup" National First Prize (2025) · ICPC China Invitational Bronze (2025)
| Venue | Project | TL;DR |
|---|---|---|
| ECCV 2026 | RoboStream 🤖 (Co-first author) | Training-free VLM planning with spatio-temporal memory for long-horizon manipulation — 90.5% success on RLBench long-horizon tasks (14.5% w/o memory) |
| ICME 2026 | Conscious Gaze 👁️ (First author) | Training-free inference-time attention intervention against VLM hallucination — POPE F1 up to +7% |
| IEEE TCC | CALO ☁️ (Student first author) | Code- & load-aware serverless resource configuration with deep RL |
SO-101 Arm ──► camera calibration · joint control · teleoperation data collection · π0 (JAX) fine-tuning
G1 / Go2 ──► PPO locomotion in MuJoCo (UniLab) · backflip · wall-jump · dancing