This is a joint project between the Planning and Diagnosis group at ANU & NICTA, and Prof. Dana Nau's lab at the University of Maryland at College Park, USA. Robots are becoming increasingly competent at sensing their environment, navigating around obstacles, and manipulating objects. At the same time, automated planning technology is becoming capable of solving a wide range of higher level planning tasks that require organising actions (such as "move from room A to room B", "pick up the blue plate", etc) into a complex plan. In order to perform household tasks such as laying a table, or assembly tasks such as assembling a piece of furniture, robots need to carry out both high level discrete planning tasks, and low level motion planning and object manipulation in continuous space. Each of these two planning types has been well researched individually, but their integration has proven challenging, largely because high level actions (such as picking up a plate) have complex geometric preconditions over continuous variables that are difficult to handle within the high level planning model.
In our group, we have developed a new planning model and algorithms that enables to couple a state of the art task planner with an external specialised "solvers" (e.g. a linear equation solver, a chemical equation solver, a temporal reasoning solver). This enables the planner to use the external solver to determine the value of certain numerical preconditions that are better determined by the specialised solver. The goal of this project is to apply and, if necessary adapt, this new planning model and the relevant algorithms to the combined task and motion planning problem.
This project is best suited to a student with a computer science or software engineering background, with a keen interest in artificial intelligence and good programming skills. It would also be suitable to a student with an engineering background and strong programming skills and knowledge of AI (e.g. some of the engineering students who took ANU COMP3620 "Artificial Intelligence" and got a D-HD mark).
(None of this is required reading before commencing the project) Task Planning Malik Ghallab, Dana Nau, and Paolo Traverso (2004). Automated Planning, Theory and Practice. Morgan Kaufmann. http://store.elsevier.com/Automated-Planning/Malik-Ghallab/isbn-9780080490519/ Hector Geffner and Blai Bonet (2013). A Concise Introduction to Models and Methods for Automated Planning. Morgan Claypool Available free from your institutional virtual library (e.g virtual.anu.edu.au) http://www.morganclaypool.com/doi/pdf/10.2200/S00513ED1V01Y201306AIM022 Motion Planning Stephen Lavalle (2006). Planning Algorithms. Cambridge university Press. Available free of charge at http://planning.cs.uiuc.edu/ Our Approach for Combining a Task Planner with a Specialised Solver Franc Ivankovic, Patrik Haslum, Sylvie Thiebaux, Vikas Shivashankar, and Dana Nau (2014). Optimal Planning with Global Numerical State Constraints. Proc. International Conference on Automated planning and Scheduling (ICAPS-14). Best Student Paper Award. http://users.cecs.anu.edu.au/~thiebaux/papers/icaps14.pdf
The student will gain a good understanding of both task and motion planning. A successful outcome would lead to the publication of a paper in an international conference. If taken as a honours/MS project, this would be a good preparation for a PhD in AI or Robotics. The student will have the possibility of interacting with partners at the University of Maryland.