Research on CSE Robotics

Robot learning and autonomous behaviour

Mon, 01 Jan 0001 00:00:00 +0000

A robot needs to adapt when its environment, task or physical behaviour changes. Learning offers ways to acquire useful behaviours from data and experience, while control connects those behaviours to the constraints of a physical system.

CSE work has combined learning from demonstration, explanation-based learning and reinforcement learning. Observations can supply an abstract account of a behaviour, which then guides trial-and-error learning. Earlier systems applied this approach to bipedal locomotion, rescue robots and learning to use objects as tools—for example, using a hook to retrieve an object beyond the robot’s reach.

Reasoning and cognitive robotics

Mon, 01 Jan 0001 00:00:00 +0000

Autonomous robots need to connect what they observe with what they know, what they want to achieve and what they can do next. Cognitive robotics studies the representations and reasoning processes that support these decisions.

Work in this theme includes reasoning about actions, task planning, knowledge representation and high-level robot programming. CSE’s earlier blackboard architectures used shared memory to connect cognitive processes. Cognitive hierarchies provided a framework for integrating symbolic and sub-symbolic representations, while action formalisms supported planning and reasoning about incomplete knowledge.

Perception and human alignment

Mon, 01 Jan 0001 00:00:00 +0000

Robots act on incomplete observations. Computer vision and multimodal perception help a robot recognise objects, interpret scenes and relate its actions to the people around it. Human–robot interaction asks how those capabilities can support understandable and useful behaviour.

This theme connects visual understanding, interactive learning and human-centred autonomy. It includes questions about how a robot can interpret feedback, communicate its intentions and adapt its behaviour to a shared task. The physical setting matters: working near a person places demands on perception, movement and the predictability of the whole system.

Multi-robot systems and embodied autonomy

Mon, 01 Jan 0001 00:00:00 +0000

A team of robots must make decisions while sharing a physical environment. Coordination depends on what each robot can observe, communicate and do, and on how those capabilities change during a task.

This theme connects multi-agent systems, collective robotics and embodied intelligence. Research questions concern how individual decisions produce useful team behaviour, how robots cope with uncertainty and how sensing and communication constrain coordination. Physical testbeds expose interactions that can be difficult to capture in a simplified model.

Software Architectures for Autonomous Robots

Mon, 01 Jan 0001 00:00:00 +0000

An example cognitive hierarchy

The CSE robotics group was an early proponent of blackboard architectures for sharing data between a robot’s cognitive processes. A blackboard provides a global shared memory facility that acts as both a communications infrastructure as well as a storage mechanism for a robot. The blackboard typically offers both publisher-subscriber based asynchronous communications as well as synchronous query-answering communications. The storage of the blackboard provide both short-term and long-term memory.

Learning Robot Behaviours

Mon, 01 Jan 0001 00:00:00 +0000

Watch the qualification video on YouTube

Demonstration of a robot learning the correct tool to use to solve a particular task.

Watch the qualification video on YouTube

Demonstration of a robot learning to climb a low-step, a high-step and staircase.

The work documented here developed machine learning techniques for robotics. Manually programming new robot behaviours is very challenging. A robot that can learn its own behaviours can better and more quickly adapt to new situations and tasks, without relying on domain experts and programmers. This learning is very beneficial for robots for home environments, as end users could instruct their robots to learn new skills, rather than rely and wait on teams of experts to design and implement new software.

Robot Control and Reasoning

Mon, 01 Jan 0001 00:00:00 +0000

Watch the qualification video on YouTube

Demonstration of a robot learning to climb a low-step, a high-step and staircase.

Our group has a strong background in formal logic based AI and particularly in the area of reasoning about knowledge and action. The expertise in this area extends from research into abstract representation formalisms for actions and knowledge, through to pragmatic issues of implementing reasoning for the purposes of controlling robot behaviour.

Position Tracking and SLAM

Mon, 01 Jan 0001 00:00:00 +0000

2D SLAM Map of the CSE Robotics Lab

Our research group has a long history of research in position tracking and SLAM, particularly for autonomous robots in the RoboCupRescue Robot competition. The work described here focused on implementing position tracking and SLAM algorithms on GPUs using full 3D information to produce correctly aligned and accurate 3D maps.

Selected publications from this work

A. Ratter, C. Sammut. Fused 2D/3D Position Tracking for Robust SLAM on Mobile Robots. International Conference on Intelligent Robots and Systems, 2015.
A. Ratter, C. Sammut. Local Map Based Graph SLAM with Hierarchical Loop Closure and Optimisation. Australasian Conference on Robotics and Automation, 2015.
A. Ratter and C. Sammut, GPU Accelerated Parallel Occupancy Voxel Based ICP for Position Tracking, 2013 Australasian Conference on Robotics and Automation, Sydney, Australia, pp. 394–403, 2013.
A. Ratter, C. Sammut, M. McGill. GPU Accelerated Graph SLAM and Occupancy Voxel Based ICP For Encoder-Free Mobile Robots. International Conference on Intelligent Robots and Systems, 2013
A. Milstein, M. McGill, T. Wiley, R. Salleh, and C. Sammut, A Method for Fast Encoder-Free Mapping in Unstructured Environments, Journal of Fields Robotics: Special Issue on Safety, Security, and Rescue Robotics, vol. 28, no. 6, pp. 817–831, 2011
A. Milstein, M. McGill, T. Wiley, R. Salleh, and C. Sammut, Occupancy voxel metric based iterative closest point for position tracking in 3D environments, IEEE International Conference on Robotics and Automation, pp. 4048 – 4053, 2011.

3D SLAM Map of the CSE Robotics Lab

General Game Playing Robots

Mon, 01 Jan 0001 00:00:00 +0000

Our research group has particular expertise in the area of General Game-Playing (GGP) and we have been active in extending this notion to the robotics domain. GGP is the attempt to create AI systems that can understand and learn to play new games without human intervention. Extending GGP to the robotics domain requires moving beyond a virtual/abstract environment to consideration of game playing within an embodied physical environment.

For example, a domestic robot tasked with fetching an item has to consider the possibility that the item may not be where it expects, or that the human operator may change locations after issuing the request. Viewing such a task as a game can provide a framework for developing more natural and intuitive robot behaviours.

Autonomous Adaptation and Trust

Mon, 01 Jan 0001 00:00:00 +0000

Trust becomes a critical issue for humans working with robots, especially when they can autonomously learn and adapt to new situations. The behaviour of these types of machines cannot be formally verified in advance. The original proposal set out to study the change in trust for a mixed initiative task under varying degrees of transparency of the adaptation process.

The proposal identified two intended contributions:

The design and development of a robotic cognitive architecture that includes the ability for the robot to adapt autonomously to a change in the task environment. The proposed architecture used a Baxter robot for participation in a mixed initiative task where the environment changes, requiring the robot to adapt on the job.
Modelling and evaluating the evolving human-robot trust relationship as the robot learns on the job.