I am an Associate Professor in Artificial intelligencee at the School of Computing and Information Systems, The University of Melbourne. I’m a member of the Agent Lab group and the Digital Agriculture, Food and Wine lab.
My research focuses on how to introduce different approaches to the problem of inference in sequential decision problems, as well as applications to autonomous systems in agriculture.
I completed my PhD at the Artificial Intelligence and Machine Learning Group, Universitat Pompeu Fabra, under the supervision of Prof. Hector Geffner. I was a research fellow for 3 years under the supervision of Prof. Peter Stuckey and Prof. Adrian Pearce, working on solving Mining Scheduling problems through automated planning, constraint programming and operations research techniques.
Graduate Certificate in University Teaching, 2020
The University of Melbourne
PhD in Artificial Intelligence, 2012
Universitat Pompeu Fabra
MEng in Artificial Intelligence, 2007
Universitat Pompeu Fabra
BSc in Computer Science, 2004
Universitat Pompeu Fabra
[8/24] New ECAI-24 Paper on Count-based Novelty Exploration in Classical Planning
[5/24] New AAMAS-24 Best-Student Paper Award on Human Goal Recognition as Bayesian Inference: Investigating the Impact of Actions, Timing, and Goal Solvability
[3/24] The University of Melbourne, Teaching Video for Algorithms and Data Structures
[2/24] New AAAI-24 Paper on Generalized Planning for the Abstraction and Reasoning Corpus
[1/24] The Guardian newspaper featured Farmbots, flavour pills and zero-gravity beer: inside the mission to grow food in space, and shared too through Farm.bot’s official newsletter
[12/23] TV Channel 10 News featured the collaborative projects on plants for Space as well as a new short video released by The University of Melbourne
[12/23] Pursuit media article You can’t explore the solar system on an empty stomach, featuring the integration of new sensors over Farm.bots
[11/23] New ICPM-23 Paper on Data-Driven Goal Recognition in Transhumeral Prostheses Using Process Mining Techniques
[10/23] New AIJ Paper on Fast and accurate data-driven goal recognition using process mining techniques
[09/23] New ECAI-23 Paper on Diverse, Top-k, and Top-Quality Planning Over Simulators
Planning as a Service (PaaS) is an extendable API to deploy planners online in local or cloud servers
Farm.bot is an open-source robotic platform to explore problems on AI and Automation (Planning, Vision, Learning) for small scale …
Width Based Planning searches for solutions through a general measure of state novelty. Performs well over black-box simulators and …
Planimation is a framework to visualise sequential solutions of planning problems specified in PDDL
Awarded top performance classical planners in serveral International Planning Competitions 2008 - 2019
Invariants, Traps, Un-reachability Certificates, and Dead-end Detection
Software to support AI courses in Mel & RMIT Unis (Melbourne, AUS)
Classical Planners playing Atari 2600 games as well as Deep Reinforcement Learning
classical planners computing infinite loopy plans, and FOND planners synthesizing controllers expressed as policies.
Lightweight Automated Planning ToolKiT (LAPKT) to build, use or extend basic to advanced Automated Planners
Width-based algorithms search for solutions through a general definition of state novelty. These algorithms have been shown to result in state-of-the-art performance in classical planning, and have been successfully applied to model-based and model-free settings where the dynamics of the problem are given through simulation engines. Width-based algorithms performance is understood theoretically through the notion of planning width, providing polynomial guarantees on their runtime and memory consumption. To facilitate synergies across research communities, this paper summarizes the area of width-based planning, and surveys current and future research directions.
Planning is the branch of Artificial Intelligence (AI) that seeks to automate reasoning about plans, most importantly the reasoning that goes into formulating a plan to achieve a given goal in a given situation. AI planning is model-based: a planning system takes as input a description (or model) of the initial situation, the actions available to change it, and the goal condition to output a plan composed of those actions that will accomplish the goal when executed from the initial situation.
The Planning Domain Definition Language (PDDL) is a formal knowledge representation language designed to express planning models. Developed by the planning research community as a means of facilitating systems comparison, it has become a de-facto standard input language of many planning systems, although it is not the only modelling language for planning. Several variants of PDDL have emerged that capture planning problems of different natures and complexities, with a focus on deterministic problems.
The purpose of this book is two-fold. First, we present a unified and current account of PDDL, covering the subsets of PDDL that express discrete, numeric, temporal, and hybrid planning. Second, we want to introduce readers to the art of modelling planning problems in this language, through educational examples that demonstrate how PDDL is used to model realistic planning problems. The book is intended for advanced students and researchers in AI who want to dive into the mechanics of AI planning, as well as those who want to be able to use AI planning systems without an in-depth explanation of the algorithms and implementation techniques they use.
Table of Contents: Praise for An Introduction to the Planning Domain Definition Language / Preface / Introduction / Discrete and Deterministic Planning / More Expressive Classical Planning / Numeric Planning / Temporal Planning / Planning with Hybrid Systems / Conclusion / Bibliography / Authors’ Biographies / Index
Count-based exploration methods are widely employed to improve the exploratory behavior of learning agents over sequential decision problems. Meanwhile, Novelty search has achieved success in Classical Planning through recording of the first, but not successive, occurrences of tuples. In order to structure the exploration, however, the number of tuples considered needs to grow exponentially as the search progresses. We propose a new novelty technique, classical count-based novelty, which aims to explore the state space with a constant number of tuples, by leveraging the frequency of each tuple’s appearance in a search tree. We then justify the mechanisms through which lower tuple counts lead the search towards novel tuples. We also introduce algorithmic contributions in the form of a trimmed open list that maintains a constant size by pruning nodes with bad novelty values. These techniques are shown to complement existing novelty heuristics when integrated in a classical solver, achieving competitive results in challenging benchmarks from recent International Planning Competitions. Moreover, adapting our solver as the frontend planner in dual configurations that utilize both memory and time thresholds demonstrates a significant increase in instance coverage, surpassing current state-of-the-art solvers, while also maintaining competitive planning time performance. Finally, we introduce two solvers implementing alternative count-based heuristics and provide promising results for future developments of the ideas presented in this study
Goal Recognition (GR) is a research problem that studies ways to infer the goal of an intelligent agent based on its observed behavior and knowledge of the environment in which the agent operates. A common assumption of GR is that the environment is static. However, in many real-world scenarios, for example, recognizing customers’ preferences, it is necessary to recognize the goals of multiple agents or multiple goals of a single agent over an extended period. Therefore, it is reasonable to expect the environment to change throughout a series of goal recognition tasks. This paper presents three process mining-based solutions to the problem of adaptive GR in a changing environment implemented as different control strategies of a system for solving standard GR problems. As a standard GR system that gets controlled, we use the system grounded in process mining techniques, as it can adjust its internal GR mechanisms based on data collected while observing the operating agents. We evaluated our control strategies over synthetic and real-world datasets. The synthetic datasets were generated using the extended version of the Goal Recognition Amidst Changing Environments (GRACE) tool. The datasets account for different types of changes and drifts in the environment. The evaluation results demonstrate a trade-off between the GR performance over time and the effort invested in adaptations of the GR mechanisms of the system, showing that few well-planned adaptations can lead to a consistently high GR performance.
Goal recognition is a fundamental cognitive process that enables individuals to infer intentions based on available cues. Current goal recognition algorithms often take only observed actions as input, but here we use a Bayesian framework to explore the role of actions, timing, and goal solvability in goal recognition. We analyze human responses to goal-recognition problems in the Sokoban domain, and find that actions are assigned most importance, but that timing and solvability also influence goal recognition in some cases, especially when actions are uninformative. We leverage these findings to develop a goal recognition model that matches human inferences more closely than do existing algorithms. Our work provides new insight into human goal recognition and takes a step towards more human-like AI models.
Planning under complex uncertainty often asks for plans that can adapt to changing future conditions. To inform plan development during this process, exploration methods have been used to explore the performance of candidate policies given uncertainties. Nevertheless, these methods hardly enable adaptation by themselves, so extra efforts are required to develop the final adaptive plans, hence compromising the overall decision-making efficiency. This paper introduces Reinforcement Learning (RL) that employs closed-loop control as a new exploration method that enables automated adaptive policy-making for planning under uncertainty. To investigate its performance, we compare RL with a widely-used exploration method, Multi-Objective Evolutionary Algorithm (MOEA), in two hypothetical problems via computational experiments. Our results indicate the complementarity of the two methods. RL makes better use of its exploration history, hence always providing higher efficiency and providing better policy robustness in the presence of parameter uncertainty. MOEA quantifies objective uncertainty in a more intuitive way, hence providing better robustness to objective uncertainty. These findings will help researchers choose appropriate methods in different applications.
In our research, we explore two orthogonal but related methodologies of solving planning instances: planning algorithms based on direct but lazy, incremental heuristic search over transition systems and planning as satisfiability. We address numerous challenges associated with solving large planning instances within practical time and memory constraints. This is particularly relevant when solving real-world problems, which often have numeric domains and resources and, therefore, have a large ground representation of the planning instance. Our first contribution is an approximate novelty search, which introduces two novel methods. The first approximates novelty via sampling and Bloom filters, and the other approximates the best-first search using an adaptive policy that decides whether to forgo the expansion of nodes in the open list. For our second work, we present an encoding of the partial order causal link (POCL) formulation of the temporal planning problems into a CP model that handles the instances with required concurrency, which cannot be solved using sequential planners. Our third significant contribution is on lifted sequential planning with lazy constraint generation, which scales very well on large instances with numeric domains and resources. Lastly, we propose a novel way of using novelty approximation as a polynomial reachability propagator, which we use to train the activity heuristics used by the CP solvers.
Jiajia Song
[2024 - current] co-supervised with Prof. Sebastian Sardina and Dr. William Umboh, Topic: What Makes AI Planning Hard? From Complexity Analysis to Algorithm Design
David Adams
[2024 - current] co-supervised with Dr. Renata Borovica-Gajic, Topic: Exploration Methods for Databases
Qingtan Shen
[2023 - current] co-supervised with A/Prof. Artem Polyvyanyy and Dr. Timotheus Kampik, Topic: Multi-agent system discovery
Muhammad Bilal
[2023 - current] co-supervised with Dr. Wafa Johal and Prof. Denny Oetomo, Topic: Towards Interactive Robot Learning for Complex Sequential Tasks
Ciao Lei
[2022 - current]. co-supervised with Dr. Kris Ehinger and A/Prof Sigfredo Fuentes, Topic: Generalized vision planning problems and their applications in Agriculture
Zhiaho Pei
[2022 - current]. co-supervized with Dr. Angela Rojas, Dr. Fjalar De Haan and Dr. Enayat A. Moallemi, Topic: Robust decision making for complex systems
Sukai Huang
[2022 - current]. co-supervized with Prof. Trevor Cohn, Topic: NLP and sequential decision problems
Lingfei Wang
[2021 - current], co-supervised with Dr.Maria Rodriguez. Topic: Scheduling and Learning for High Performance Computing (HPC)
Guang Hu
[2020 - current], co-supervised with Dr.Tim Miller. Topic: Epistemic Planning and Explanability
Zihang Su
[2020 - current], co-supervised with Dr. Artem Polyvyanyy and Prof. Sebastian Sardina. Topic: Bridging the gap between Process Mining and Goal-Plan-Intention Recognition
Chenyuan Zhang
[2020 - 2024], co-supervised with A/Prof. Charles Kemp (Psychology). Thesis:
Planning and Goal Recognition in Humans and Machines First Employment: Post-Doc @ Monash University [2024 - current] Best Student Paper Award AAMAS (2024)
Anubhav Singh
[2019 - 2024], co-supervized with Dr. Miquel Ramirez and Prof. Peter Stuckey. Thesis:
Lazy Constraint Generation and Tractable Approximations for Large-scale Planning Problems First Employment: Post-Doc @ Universtiy of Toronto [2024 - current]
Stefan O'Toole
[2018 - 2022], co-supervized with Dr. Miquel Ramirez and Prof. Adrian Pearce. Thesis:
The Intersection of Planning and Learning through Cost-to-go Approximations, Imitation and Symbolic Regression First Employment: Meta [2022 - current]
Toby Davies
, [2013-2017], co-supervized with Prof. Adrian Pearce, Prof. Peter Stuckey and Prof. Harald Sondergaard. Thesis:
Learning from Conflict in Multi-Agent, Classical, and Temporal Planning. First Employment: Google [2017 - current]. Best Paper Award ICAPS (2015)
, Best PhD Thesis, Melbourne School of Engineering 2018
Giacomo Rosa
[2023-2024]. Thesis:
Regression and Width in Classical Planning and
ECAI24 paper
Zhiaho Pei
[2021]. co-supervized with Dr. Angela Rojas, Dr. Fjalar De Haan and Dr. Enayat A. Moallemi, Thesis:
Robust decision making for complex systems
Marco Marasco
[2021]. co-supervized with Dr. Angela Rojas, Dr. Fjalar De Haan and Dr. Enayat A. Moallemi, Thesis:
Adaptive Policy making for systems of electricity provision
Jiayuan Chang
[2021]. co-supervized with A/Prof Sigfredo Fuentes, Thesis:
FarmBot.io Automated Planning: simulation and integration
Yajing Ma
[2021]. co-supervized with A/Prof Sigfredo Fuentes, Thesis:
Electronic Nose for pest detection
Dmitry Grebenyuk
[2018-2020], co-supervised with Dr. Miquel Ramirez, and Dr. Kris Ehinger. Thesis:
Agnostic Features for generalized policies computed with Deep Reinforcement Learning (DRL). First Employment: Start-up working on Image Processing using DRL
Guang Hu
[2018-2020], co-supervised with Dr.Tim Miller. Thesis:
What you get is what you see: Decomposing Epistemic Planning using Functional STRIPS. PhD Candidate [2020 - current]
Ciao Lei
[2019-2020]. Thesis:
Regression and Width in Classical Planning and
ICAPS21 paper
ICAPS (2025)
ICAPS (2019)
International Conference on Automated Planning and Scheduling – Publicity co-chair, ICAPS (2010)
First Unsolvability International Planning Competition – Co-Organizer, UIPC-1 (2016)
Heuristics and Search for Domain-independent Planning – Co-Organizer, ICAPS workshop HSDIP (2015,2016,2017,2018)
Demonstration track – Co-Chair AAAI 2023
Student Abstract track – Co-Chair, AAAI (2018,2019)
Journal Presentation track – Co-Chair ICAPS (2018)
AAAI (2020,2021,2022,2023)
IJCAI (2021,2023)
International Joint Conferences on Artificial Intelligence IJCAI (2011,2013,2015,2017,2018,2020,2022)
Association for the Advancement of Artificial Intelligence, AAAI (2013,2015,2016,2017,2018,2019)
European Conference on Artificial Intelligence, ECAI (2014,2016)
International Conference on Automated Planning and Scheduling, ICAPS (2015,2016,2017,2018,2020)
Symposium on Combinatorial Search SOCS (2020,2021,2022,2023)
Journal of Artificial Intelligence Research, JAIR
Reviewer Artificial Intelligence, Elsevier AIJ
Pacman Capture the flag Inter-University Contest
, run for Unimelb AI coure and
Hall of Fame contest, 2016 - current
AI Planning for Autonomy (Lecturer)
, at M.Sc. AI specialization, The University of Melbourne,
2016 - current
Data Structures and Algorithms (Lecturer)
, at The University of Melbourne,
2016 - current
Software Agents (Lecturer)
, at M.Sc. Software, The University of Melbourne, 2013, 2014, 2015
Autonomous Systems
, at M.Sc. Intelligent Interactive Systems, University Pompeu Fabra, 2012
Advanced course on AI: workshop on RoboSoccer simulator
, at Polytechnic School, University Pompeu Fabra, 2009, 2010, 2011
Artificial Intelligence course
, at Polytechnic School, University Pompeu Fabra, 2010, 2011
Introduction to Data Structures and Algorithms course
, at Polytechnic School, University Pompeu Fabra, 2008
Programming course
, at Polytechnic School, University Pompeu Fabra, 2008, 2009, 2010, 2011