My research interests lie in the Artificial Intelligence areas of data mining, recommender systems, multi-agent systems, and autonomous robotics.
Previous to this, my PhD research focused on distributed multi-agent task clustering and allocation in a market-based environment.
For information about these works please visit my university hosted personal webpage.
PhD Research – Sequential Single Cluster Auctions
I completed my PhD in the Artificial Intelligence Group, in the School of Computer Science and Engineering at the University of New South Wales from 2010 – 2013. During this time I published five papers, gave four conference presentations, one poster presentation, and a doctoral consortium presentation.
You can download the thesis here.
This thesis studies task allocation in multi-robot teams operating in dynamic environments. The multi-robot task allocation problem is a complex NP-Complete optimisation problem with globally optimal solutions often difﬁcult to ﬁnd. Because of this, the rapid generation of near optimal solutions to the problem that minimise task execution time and/or energy used by robots is highly desired. Our approach seeks to cluster together closely related tasks and then builds on existing distributed market-based auction architectures for distributing these sets of tasks among several autonomous robots.
Dynamic environments introduce many challenges that are not found in closed systems. For instance, it is common for additional tasks to be inserted into a system after an initial solution to the task allocation problem is determined. Additionally, it is highly likely in long-term autonomous systems that individual robots may suffer some form of failure. The ability to alter plans to react to these types of challenges in a dynamic environment is required for the completion of all tasks. In our approach we allow the repeated formation and auctioning of task clusters with varying tasks. This allows us to react to and change the task allocation among robots during execution.
Throughout this thesis we use empirical evaluation to study different approaches for forming clusters of tasks and the application of task clustering to distributed auctions for multi-robot task allocation problems. Our results show that allocating clusters of tasks to robots in solving these types of problems is a fast and effective method and produces near optimal solutions.
- B. Heap and M. Pagnucco. Repeated Auctions for Reallocation of Tasks with Pickup and Delivery Upon Robot Failure. In Proceedings of the 16th International Conference on Principles and Practice of Multi-Agent Systems (PRIMA), LNCS 8291, 2013.
- B. Heap and M. Pagnucco. Repeated Sequential Single-Cluster Auctions with Dynamic Tasks for Multi-Robot Task Allocation with Pickup and Delivery. In Proceedings of the 11th German Conference on Multiagent System Technologies (MATES), LNCS 8076, 2013.
- B. Heap and M. Pagnucco. Analysis of Cluster Formation Techniques for Multi-Robot Task Allocation using Sequential Single-Cluster Auctions. In Proceedings of the 25th Australasian Joint Conference on Artificial Intelligence (AI12), LNCS 7691, 2012.
- B. Heap and M. Pagnucco. Repeated Sequential Auctions with Dynamic Task Clusters. In Proceedings of the Twenty-Sixth AAAI Conference on Artificial Intelligence (AAAI-12), 2012.
- B. Heap and M. Pagnucco. Sequential Single-Cluster Auctions for Robot Task Allocation. In Proceedings of the 24th Australasian Joint Conference on Artificial Intelligence (AI11), LNCS 7106, 2011.
My research interests at Massey University were in agent based simulation and artificial intelligence. My research at both honours and undergraduate levels has focused on agent based vehicle traffic simulation. At Massey University I was a member of the Complex Systems and Simulations Research Group within the Institute of Information and Mathematical Sciences.
Honours Research Project
My honours project for my BSc (Hons) focused on the modeling of congestion within a traffic network. The ideas for the project built upon the ideas explored in my undergraduate research project. The final project was completed in Java and used Java OpenGL (JOGL) to render the 3D environment.
The simulation of traffic involves the modeling of a complex system that is open, shows emergent phenomena, and non-linear relationships. This project outlines the ideas involved in traffic simulation and reports on the results of a traffic simulator designed to model congestion amongst urban and suburban roads. The traffic simulator in this project uses artificial intelligence to give individual behavioural characteristics to each vehicle within the simulation and successfully model their interactions between vehicles and the environment. Overall this project successfully shows a relationship between the amount of congestion present in a traffic network and the mean speed of the vehicles in the network, and the effects of various different intersection controllers on traffic flow. It is discovered that the best method for maintaining high mean speeds as traffic flows through intersections is to use a mixture of intersection controllers on connected roads to vary the flow of traffic on each road.
- Download Thesis
- Download Simulator JAR File (you will need JOGL installed to run)
- Download Source Code
Undergraduate Research Project: Traffic Simulation
This project was completed in Semester Two 2007 during my final year of my BSc. The project was completed by me and fellow student Dean Jerkovich with supervision from Professor Ken Hawick. Both I and Dean were awarded an A+ grade for our work.
This project was my first attempt at building a simulator. At the end of the project we were able to successfully model multiple agents and their interactions with each other. The simulator was built in Java and allowed the user to design a traffic network consisting of roads with multiple lanes and intersections which were controlled by simple traffic lights. Cars are generated at random locations on each road at the start of the simulation and will have a randomised goal location for each of them to reach. The cars have no artificial intelligence for navigation so as a result they drive randomly around the network until they stumble upon their goal. The simulator stops running when all cars have located their goal.
Each of the cars in the simulator has artificial intelligence controlling their acceleration, braking, awareness of vehicles in front, and overtaking. Each agent also has a randomly distributed normal variable which determines how far below or above the speed limit the vehicle will travel at.