Hello

I’m Xiawei (William) Liu, an undergraduate focused on robotics, trustworthy ML, and reliable systems.

I study Computer Engineering & Computer Science at USC, supervised by Prof. Yue Wang. I work on humanoid manipulation datasets, teleoperation systems, and evaluation pipelines that make research reproducible and explainable for new collaborators.

Los Angeles, CA Graduating 2026 Open to internships & research roles
Portrait of Xiawei (William) Liu
Xiawei (William) Liu

Publications

Selected papers and preprints.

Humanoid Everyday: A Comprehensive Robotic Dataset

ICRA (under review), 2025

Zhenyu Zhao*, Hongyi Jing*, Xiawei Liu, et al. “Humanoid Everyday: A Comprehensive Robotic Dataset for Open-World Humanoid Manipulation.”

arXiv:2510.08807
Citation (BibTeX)
@misc{zhao2025humanoideverydaycomprehensiverobotic,
  title={Humanoid Everyday: A Comprehensive Robotic Dataset for Open-World Humanoid Manipulation},
  author={Zhenyu Zhao and Hongyi Jing and Xiawei Liu and Jiageng Mao and Abha Jha and Hanwen Yang and Rong Xue and Sergey Zakharor and Vitor Guizilini and Yue Wang},
  year={2025},
  eprint={2510.08807},
  archivePrefix={arXiv},
  primaryClass={cs.RO},
  url={https://arxiv.org/abs/2510.08807},
}

Research experience

Hands-on work with labs and collaborators.

Humanoid Everyday · Physical SuperIntelligence Lab (PSI Lab), USC

2025 — Present
  • Designed and collected multi-modal humanoid datasets (RGB, depth, LiDAR, tactile, IMU, joint states, language) at 30 Hz on Unitree G1/H1.
  • Engineered a teleoperation pipeline (Apple Vision Pro → hands/arms via Pinocchio IK, shared-memory I/O), cutting control delay from ~500 ms to ~2 ms.
  • Built cloud evaluation for remote policy upload and hardware-in-the-loop testing; sustained 100+ minutes of continuous trials.
  • Benchmarked policies (DP/DP3, ACT, OpenVLA, π₀-FAST, π₀.5, GR00T N1.5) and analyzed failure modes.

Whole-Body Foundation Model Evaluation · Physical SuperIntelligence Lab (PSI Lab), USC

2025
  • Built a Viser-based visualization and data-verification tool to inspect trajectories and diagnose end-effector pose discrepancies.
  • Optimized evaluation pipeline initialization from 2 minutes to 2 seconds, increasing experiment throughput.

Alias-Free GAN Image Detection · CMU

2024
  • Ran double-blind studies on StyleGAN3 faces (FFHQ) and animals (AFHQv2) at truncation 0.7; analyzed misclassification patterns.
  • Generated controlled 1024×1024 images; built a rubric for realism, coherence, and detail; anonymized datasets and managed evaluator workflow.
  • Implemented analysis pipeline (Python; statistics, matplotlib) and co-authored methodology and discussion.

Projects

Academic and systems work.

Network Attached Storage (NAS)

Raspberry Pi
  • Personal NAS with remote WAN access via intranet penetration and port forwarding.
  • VPN setup for secure WAN access.

Embedded Systems Development

Embedded C / Python
  • IoT hardware interfacing (sensors, actuators) and distributed communication.
  • Arduino Uno speed-trap system as a representative build.

Bitrate Adaptation in an HTTP Proxy

C++
  • HTTP proxy that adjusts video streaming quality based on network conditions.

Custom DNS Load Balancer

C++
  • Round-robin and geographic-based DNS nameserver for load balancing.

MQTT & JSON Serialization

Networking
  • Data transmission system using MQTT with JSON serialization.

Algorithm Implementations

C++
  • AVL Tree, Dijkstra’s, A*, and sorting algorithms (Merge, Quick, Heap).