Professor Sylvie Thiébaux and Associate Professor Charles Gretton.
Even as AI tools and large language models ChatGPT and Claude spread throughout the global economy, much of our modern world runs on automated planning and scheduling, ranging from airports to warehouse management.
The Australian National University School of Computing hosts one of the largest AI planning groups in the world, a critical mass of researchers solving core challenges at the heart of global AI research. AI planning allows machines to think ahead, make decisions, and adapt to changing conditions. For example, a drone needs to quickly develop a strategy to avoid crashing into an unknown object.
“It’s a huge open problem,” Associate Professor Charles Gretton said. “ANU is lucky to have some of the best researchers in AI planning.”
One of those researchers, Professor Hanna Kurniawati, looks at sequential decision-making for robotics, especially how to make them perform well despite various uncertainties.
“There’s no such thing as perfect sensors,” Kurniawati explained. “And there are uncertainty errors in the robot’s understanding of how the world behaves. It might think that behind a door there are problems, wondering whether it is a place the robot can go or not.”
A key element of the interdisciplinary AI ecosystem at the ANU
AI Planning researchers from ANU attending ICAPS 2025.
Many of the planning researchers at ANU are involved in the Integrated AI Network, which brings together experts from different disciplines across ANU whose work focuses on the role of AI in addressing social and scientific problems, as well as utilising AI in creative spaces and cultural practices. Projects range from plasma physics to data-driven health and welfare research and serve to centre the university’s role as a leading AI research institution.
“Australia is a resource and agricultural economy, and the big value proposition is physical AI and remote autonomy,” Gretton said. “Looking after ecology, automated mining, reducing the mining footprint. We need machines to use all the data they collect and deliberate in a safe and trustworthy way. Planning is the technology behind that.”
Approaches used in planning are typically based on reasoning or combinations of reasoning and learning. This distinguishes them from pure machine learning approaches like neural networks, which often have a “black box” problem when it comes to explaining why the AI made a specific choice.
“AI that plans or reasons deliberately has to navigate a sequence of choices, actions, and consequences,” said Senior Lecturer and robotics expert Rahul Shome. “It can explain why the AI does X but not Y. This contrasts with other forms of AI like LLMs, which tend to be less transparent and explainable.”
Record attendance and leadership at premiere planning conference ICAPS
Professor Hanna Kurniawati delivering her keynote address. (Source: LinkedIn)
Nineteen ANU faculty and students attended 2025’s International Conference on Automated Planning and Scheduling. The conference focuses on the theory and applications of intelligent and automated planning and scheduling technology. Held in Melbourne alongside two other conferences, the International Conference on Principles of Knowledge Representation and Reasoning (KR) and the International Conference on the Integration of Constraint Programming, Artificial Intelligence, and Operations Research (CPAIOR), there were several hundred total attendees, many of whom are global leaders in classical AI and AI planning.
ANU researchers gave two conference keynotes and led three workshops between the three conferences. Kurniawati’s ICAPS-CPAIOR joint keynote, “Sequential Decision-Making for Robots Operating in Non-Deterministic and Partially Observable Worlds,” explored the question of how a robot should strategize to achieve its goals despite various types of uncertainty, while Professor Sylvie Thiébaux’s keynote for KR, “Graph Learning for Planning,” discussed advanced techniques for representing planning problems using graph neural networks.
ICAPS started as a merger of European and American precursors which date back the 1990s. While some academic conferences can feel overwhelmingly large, ICAPS provides a tight-knit and welcoming environment for longtime attendees and early-career researchers. Thiébaux attended ICAPS precursors starting in 1992 as a master’s student and cites the community as a reason to keep coming back.
“I’ve seen this community evolve over the years,” Thiébaux said. “The best work in planning you put in ICAPS rather than these big conferences. We’re very welcoming of new people.”
Thiébaux became president of ICAPS’ council from 2010-2012 and organized the 2007 and 2022 conferences. Gretton, one of her former PhD students, served as the sponsorship chair for the 2025 conference. In this role, Gretton engaged with stakeholders throughout the ICAPS community, from PhD students, scholarship sponsors, local business partnerships, and summer school programming for early-career researchers. 2025 partners included Google, Learning Machine, J.P. Morgan Chase, and optimisation research centre OPTIMA.
Supporting early-career researchers
Associate Professor Pascal Bercher and PhD student Pascal Lauer.
ICAPS supports a doctoral consortium that provides a space for PhD students to present their research and receive feedback, as well as a summer school program for PhD students held in advance of the conference, featuring lectures by senior ICAPS figures.
“By the time you arrive at the conference, you’re going to know your fellow students, several researchers, and you’ll be comfortable,” Thiébaux added. “Conferences are for both learning and networking. Be yourself and don’t censor yourself. When I was young, I was quite outspoken and I believe this helped me.”
Associate Professor Pascal Bercher is another accomplished AI planning researcher at the ANU. His research focuses on hierarchical planning, an approach to automated planning in which the dependency among actions can be given in the form of hierarchically structured networks. Bercher gave a workshop keynote for aspiring PhD students at the conference on how to plan their path through academia.
“It’s specifically designed for students who are not yet PhD students,” Bercher explained. “I wanted to share my life lessons – one of which is to decide what you really want to do and whether you want to go down that path for the right reasons. You need a great supervisor, someone who has time for you and is right for your research discipline. I genuinely like mentoring; in the time I need to create a paper myself, I can supervise four.”
Pascal Lauer, one of Bercher’s own PhD students, attended ICAPS 2025 and had a paper accepted for ICAPS 2026, to be held in Dublin, Ireland.
“What I like about working with [Bercher] is that he’s really quite encouraging,” Lauer said. “I really like ICAPS because it feels like a family. Everyone is interested in what you’re doing, and there’s no real barrier to ask questions of the greatest researchers in the field.”