Changlin Song 宋昌霖

PhD student in Computing at the Australian National University, supervised by Dylan Campbell. I work on embodied agents and control, computer vision, 3D vision and spatial intelligence.

Portrait of Changlin Song

News

Research

Boston Dynamics Spot rolling a blue water jug on grass

The agent authors the skill. Physics runs it.

Changlin Song, with GE and the University of Pennsylvania

Ongoing project, 2026

project page

An LLM agent writes whole-body skills for a Boston Dynamics Spot, and a physics-based sampling MPC runs them outdoors with no demonstrations and no per-task training.

Qualitative results of extreme amodal face detection

Extreme Amodal Face Detection

Changlin Song, Yunzhong Hou, Michael Randall Barnes, Rahul Shome, Dylan Campbell

WACV, 2026  (Oral)

project pagearXiv

Detecting faces that are partly or entirely out of frame.

FlashDreamer pipeline figure

Enhancing Monocular 3D Scene Completion with Diffusion Model

Changlin Song, Jiaqi Wang, Liyun Zhu, He Weng

arXiv, 2025

arXivcode

Completing a 3D scene from a single image without training.

FedDistill method figure

FedDistill: Global Model Distillation for Local Model De-Biasing in Non-IID Federated Learning

Changlin Song, Divya Saxena, Jiannong Cao, Yuqing Zhao

arXiv, 2024

arXiv

De-biasing local training in federated learning with group distillation.

Education

Teaching and service