Ziyi Chang

From Kinematics to Physics and Embodiment

I study human motion at multiple levels of realism, from kinematic motion synthesis to physics-aware interaction and embodied agents. This connects visually plausible animation with contact-rich interaction, physical constraints, and agents that can act within shared environments.

From Individuals to Interactions and Crowds

My work expands motion modelling from individual characters to dense interaction and crowd-level behaviour. I am interested in how local motion, pairwise or group interaction, and large-scale collective patterns can be represented within a coherent modelling framework.

From Generation to Understanding and Intervention

I view generative modelling as one part of a broader reasoning pipeline. The goal is not only to synthesize plausible motion, but also to understand interaction patterns and support intervention, planning, or analysis in dynamic human-centred environments.

Trustworthy, Controllable, and Interpretable Deployment

Across these axes, I aim to make motion intelligence robust, controllable, and interpretable. Building on my work on adversarial attacks and trustworthy AI, I study how motion models behave under perturbations, ambiguous observations, and safety-critical deployment conditions.

GROWD: Generative Modelling of Dense Crowds under Multiscale Observations, 2027 - 2028

This Bienvenüe+ MSCA-COFUND Fellowship supports a 24-month independent research programme that I proposed and developed on generative modelling of dense crowds under multiscale observations.

• Role: Project Lead / Research Fellow
• Host Institution: Inria Rennes
• Host Team: VirtUs
• Duration: 24 months
• Project Acronym: GROWD
• Research Direction: Dense crowd modelling, generative models, and crowd dynamics
• Core Aim: Integrating sparse, local, fine-grained agent-level observations with dense, global, coarse-grained video-derived measurements for uncertainty-aware crowd analysis.

View the GROWD project overview

Cross-Institutional Research Capacity Development in Human-Robot Interaction, 2025 - 2026

• Funding Source: Singapore Management University / Durham University, Singapore / UK
• Reference Number: 3787041; Value: £15,000
• Role: Research Assistant / Project Contributor
• Responsibilities included reviewing diffusion models in human-robot interaction, deploying trajectory planning experiments, managing the progress of another research assistant, and communicating with the SMU team on behalf of the Durham side.

View the SMU-DU project overview

GROWD: Generative Modelling of Dense Crowds under Multiscale Observations

GROWD is my Bienvenüe+ MSCA-COFUND Fellowship project, hosted by the VirtUs team at Inria Rennes. The project develops probabilistic generative models for dense crowd dynamics by combining sparse, local, fine-grained agent-level observations with dense, global, coarse-grained video-derived measurements.

The project addresses a key limitation in current crowd modelling: existing approaches often either assume unrealistic access to complete individual trajectories or rely on coarse field-level observations that obscure local interactions. GROWD treats these two sources of information as complementary views of the same underlying crowd dynamics.

The project is supported by the Bienvenüe+ MSCA-COFUND Fellowship, co-funded by the European Union's Horizon Europe Marie Skłodowska-Curie Actions COFUND programme, Région Bretagne, and participating host institutions.

Cross-Institutional Research Capacity Development in Human-Robot Interaction

This SMU-Durham project supported cross-institutional research capacity development in human-robot interaction, connecting diffusion-model research, trajectory planning experiments, and collaborative project coordination between Singapore Management University and Durham University.

My role involved reviewing diffusion models for human-robot interaction, deploying trajectory planning experiments, managing the progress of another research assistant, and communicating with the SMU team on behalf of the Durham side.

The project connects to my broader research interests in generative motion modelling, human-robot interaction, and trustworthy embodied AI.

Final Year Project: Styled Walking Synthesis with Diffusion Models

This undergraduate final-year project explored denoising diffusion probabilistic models for stylized human walking synthesis. I co-supervised Edmund with Prof. Hubert P. H. Shum, supporting the project direction, experiment design, and paper development.

The project led to the paper Denoising Diffusion Probabilistic Models for Styled Walking Synthesis, and forms an early link between my supervision experience and my research on generative human motion.

PhD Co-Supervision

Co-supervising a PhD student with Prof. Hubert P. H. Shum at Durham University on interactive motion modelling and character animation.

• Providing research guidance on problem formulation, experimental design, paper writing, and revision.
• Co-supervised work has led to accepted papers at Eurographics 2026 and SIGGRAPH Asia 2025, including Physics-Based Motion Tracking of Contact-Rich Interacting Characters and Motion In-Betweening for Densely Interacting Characters.
• This supervision experience directly informs my broader research agenda on multi-character interaction and socially intelligent embodied agents.

Final Year Project Supervision

Co-supervised undergraduate student Edmund with Prof. Hubert P. H. Shum on stylized human motion generation.

• The project led to the paper Denoising Diffusion Probabilistic Models for Styled Walking Synthesis.
• The work connects undergraduate research supervision with my broader interests in generative human motion and character animation.

View the final-year project overview

University Teaching

Delivered practical and workshop sessions for undergraduate and postgraduate modules at Durham University.

• Worked in teaching teams with lecturers, reporting to Dr Barnaby Martin.
• Teaching areas include programming, data science, computational thinking, data analytics, and machine learning.
• Supported mixed-background cohorts through hands-on sessions, debugging support, assessment preparation, and project guidance.

Conference Organisation

Supporting Chair of the 21st ACM SIGGRAPH / Eurographics Symposium on Computer Animation (SCA 2022).

• Led a 7-person support team.
• Coordinated with conference chairs, industrial partners, academic participants, business contacts, and service suppliers.

Outreach and Public Engagement

Co-designed and delivered robotics and AI workshops for children at Bullion Hall, UK.

• Promoted inclusive participation in STEM, with particular attention to encouraging girls to explore robotics and AI.
• Delivered accessible AI education sessions for older adults, focusing on practical understanding and trust in everyday AI technologies.

Invited Talks and Seminars

Coming soon.