ICCV 2021 Mair2 Workshop

Multi-Agent Interaction and Relational Reasoning

First Edition

October 17, 2021 | Montreal, Canada (Virtual)

Welcome to ICCV MAIR2 Workshop

Modeling relations and interactions between agents (e.g., objects, robots, and humans) is widely studied and plays an important role in various tasks, necessitating larger-scale communication and collaboration between researchers in different fields. Our goal is to enable interdisciplinary discussion about multi-agent relational reasoning from different research areas such as autonomous driving, visual reasoning, multi-agent systems, object detection and tracking, scene understanding, human-robot interaction, graph representation learning, intuitive physics, dynamics modeling, and cognitive science. The goal of this workshop is to provide an opportunity to discuss how the concepts and techniques in different fields related to multi-agent interaction modeling could help advance each other.

Workshop Recording


  • October 17, Our workshop is happening now! Please join using the above login information.

  • April 16, Paper submission opens. We feature three tracks: (1) Non-proceeding short papers (≤ 4 pages); (2) non-proceeding regular papers; (3) regular papers in proceedings

  • April 12, Our workshop website is online

  • April 08, Our workshop is accepted at ICCV 2021!

Subscribe to our mailing list

Topics Covered

  • Multi-agent datasets with complex interactions and relations (e.g., autonomous driving datasets, multi-robot datasets)

  • Relational reasoning in perception (e.g., detection, tracking) and scene understanding (e.g., segmentation, scene graph)

  • Relational reasoning and interaction modeling in intuitive physics and dynamics

  • Interaction modeling in trajectory forecasting of traffic participants (pedestrians and vehicles), sports players, etc.

  • Multi-agent motion planning and decision making

  • Visual transformers and their applications to relational reasoning

  • Relational representation learning and graph neural networks

  • Spatio-temporal relational reasoning

  • Human-robot / human-object / robot-robot interaction, human-robot collaboration

  • Multi-robot systems and swarm systems

  • Cooperative and competitive multi-agent systems

  • Causal discovery and modeling

  • Evaluation and metrics of interaction modeling

  • Transfer learning and domain adaptation in relational modeling

  • Explainability and interpretability in relational reasoning and interaction modeling

  • Multi-agent reinforcement learning with interaction modeling

Invited Speakers

Jiajun Wu

Assistant Professor

Stanford University

Alexander Schwing

Assistant Professor


Micol Marchetti-Bowick

Tech Lead Manager


Han Hu

Principal Researcher

Microsoft Research Asia