Presenter
Ass Prof Hamid Rezatofighi
Department of Data Science and AI
Monash University
Associate Professor Hamid Rezatofighi is an AI and robotics researcher in the Faculty of Information Technology at Monash University, Australia. His research spans computer vision, machine learning, robotics, and neuro-symbolic AI, with a particular focus on building intelligent autonomous systems capable of perception, reasoning, planning, and reliable decision-making in complex and dynamic environments. His current research explores combining neural foundation models with symbolic reasoning and agentic planning to develop more capable, interpretable, and trustworthy embodied AI systems, with applications including autonomous systems, human-centred robotics, and Defence.
Dr. Rezatofighi has authored more than 120 publications in top-tier venues across computer vision, machine learning, artificial intelligence, and robotics, attracting over 21,000 citations. His work on Generalized Intersection over Union (GIoU) has received over 9,300 citations and has been widely adopted in modern object detection systems.
He has attracted more than $18 million in competitive research funding, including multiple DARPA projects, an ARC Discovery Project as Lead CI, and the U.S. Office of Naval Research Global-X Challenge Award as Lead PI. His research achievements have also been recognised through the Australian Government Endeavour Research Fellowship, undertaken at Stanford University’s Vision and Learning Lab, and the Monash Faculty of IT Dean’s Award for Early Career Research Excellence.
He actively serves the international AI research community through senior editorial roles and, since 2020, regular service as Area Chair or Lead Area Chair for major international AI, machine learning, and computer vision conferences.
Ass Prof Hamid Rezatofighi of Monash University, will present a seminar on Thursday, 27 August 2026.
Title: Neuro-Symbolic AI for Visual Reasoning and Robotic Systems: Building Trustworthy Intelligence.
Abstract: Robots and autonomous systems are rapidly becoming part of complex real-world environments, driven by remarkable advances in neural foundation models. Yet impressive demonstrations of robotic skills do not necessarily translate into intelligent and trustworthy autonomy. Today, much of industry and academia focuses on scaling neural foundation models with larger models, more data, and increased computation—approaches that deliver impressive capabilities but often remain difficult to interpret, struggle with complex reasoning, and make safety assurance challenging. These limitations become particularly important in Defence and other mission-critical environments, where autonomous systems must operate reliably under uncertainty, incomplete observations, dynamic conditions, and operational constraints.
In this talk, I will present our work at Monash exploring a complementary direction: neuro-symbolic AI for building autonomous systems that are more intelligent, interpretable, and safer. Rather than replacing neural foundation models, our approach combines their strengths in perception and language understanding with explicit world representations, reasoning, and planning. This enables autonomous systems not only to perceive and act, but also to maintain knowledge about their environment, reason about complex missions, explain and ground their decisions, and provide greater opportunities for human oversight and safety assurance.
I will first showcase how we are extending these principles to humanoid and human-centred robots, where persistent world understanding, reasoning about people and their relationships, natural human–robot communication, and safe interaction become increasingly important.
I will then demonstrate these ideas through our Defence-oriented research in the DARPA Assured Neuro-Symbolic Reasoning program, where we developed an end-to-end autonomous UAV system capable of understanding high-level mission instructions, actively searching complex environments, reasoning over accumulating evidence, and planning its actions under operational constraints. The system achieved leading performance in independent program evaluations and progressed to real-world UAV demonstrations.
Finally, I will highlight findings from our recent DARPA- and ONR-supported research in visual reasoning, exposing important limitations of current foundation models and motivating approaches that go beyond purely end-to-end neural scaling. These capabilities are relevant to applications such as autonomous search, surveillance and reconnaissance, search and rescue, human–robot teaming, and decision support in complex environments.
Together, these examples point toward a future in which neural learning, structured reasoning, and planning work together to build trustworthy intelligence for real-world autonomous systems.
Click the button below to register, add an invitation to your calendar and join the seminar using the Teams/GovTeams link.
