News

20/01/09 - Research Fellowship available

A research fellowship is available on the new CogX project to work on architectures, motivation and integration for cognitive robotics. If you are interested in applying read more about the job.

---

Grant Success: Automated Diagnosis for Autonomous Underwater Vehicles

We are delighted to announce another grant success. Richard Dearden will lead a project on Automated Diagnosis for Fault Detection, Identification and Recovery in Autosub6000. This project will help make autonomous underwater vehicles (AUVs) more reliable by enabling them to deal with potentially dangerous faults they may have. This will be applied to the Autosub programme in the Underwater Systems Lab (USL) at the National Oceanography Centre in Southampton . This series of AUVs is used to help us understand the deep ocean and climate change. The project will last three years, and is 320k pounds in value. We gratefully acknowledge funding from NERC (Natural Environment Research Council) under their Oceans 2025 initiative.

---

PhD Student wins internship at Monterey Bay Aquarium Research Institute.

Zeyn Saigol , a PhD student in the Intelligent Robotics Lab has won an internship to spend the summer at the Monterey Bay Aquarium Research Institute (MBARI) . Zeyn works on AI planning techniques for autonomous underwater vehicles. MBARI is one of the leading oceangraphic research institutes in the world, with an excellent programme in autonomy, to which Zeyn will be attached. Zeyn was one of 10 interns chosen from 170 applicants. Enjoy the Californian sunshine Zeyn!

---

Overview of the group

We conduct research in a number of sub-fields of intelligent robotics and related areas such as statistical machine learning. At the most general level members of the lab use computational models, implemented in robots to test ideas about how intelligence can and should work. Some lab members work with those studying biological intelligence. Others work on the engineering science challenge of creating new algorithms required for robots to learn, see, communicate and plan. Finally we work on how to put all these together in complete systems.

This grouping covers research on architectures for accounting for human mental states and processes as well as recreating them in computer programs. Analysing architectures for human mental states and processes allows us to investigate, for instance, whether the ability to have emotional states is an accident of animal evolution or an inevitable consequence of design requirements and constraints.

The lab currently holds external research funding on projects either due to start, or in progress worth in excess of 2.2million pounds. We gratefully acknowledge the support of all our funders past and present. These include the European Commission IST Directorate Cognition Unit, the Leverhulme Trust, the Engineering and Physical Sciences Research Council (EPSRC), the Natural Environment Research Council (NERC). Past funders include the Royal Society, Advantage West Midlands and the British Council. Please look at the following project summaries, and view our collection of movies of robots in the lab.

Projects

CoSy: Cognitive Systems for Cognitive Assistants

The CoSy project is an interdisciplinary four year project on cognitive robotics funded by the European Commission, involving seven universities. At Birmingham we are working on cognitive architectures, learning about human actions, and learning about affordances and object behaviour.

Cognition and Affect

The main goal of this project is to understand the types of architectures that are capable of accounting for the whole range of human (and non-human) mental states and processes, including not only intelligent capabilities, such as the ability to learn to find your way in an unfamiliar town and the ability to think about infinite sets, but also moods, emotions, desires, and the like.

CogX: Cognitive Systems that Self-Understand and Self-Extend

This project is concerned with building a theory, together with robotic implementations, of how cognitive systems, particularly robots, can understand what they know, and how what they know changes when they act. The IR Lab will also be the coordinator of this project, which starts May 1st 2008 and runs until the end of June 2012. Details are available on the local project web site . The project is led by Jeremy Wyatt, Richard Dearden, Aaron Sloman and Nick Hawes.

Learning in Committee Machines

In the early parts of this research we explained why committees of learning machines are better if there is diversity in the errors made by committee members. This is a long standing problem in ensemble learning, and our work led to a prize winning dissertation. This work is now continuing in the EPSRC funded ADEPT project. ADEPT stands for Adaptive Dynamic Ensemble Prediction Techniques, and is lead by Dr Gavin Brown of Manchester University and Dr Tim Kovacs of the University of Bristol. The project will last for three years and is concerned with understanding the relationship between two different types of machine learning techniques: learning classifier systems and committee machines.

Planning, learning and sequential decision making

How should an agent decide how to act in an uncertain world? There are many kinds of uncertainty, perhaps the effects of actions are uncertain, or observations are noisy or aliased, maybe either or both of these are both uncertain and also unknown. We typically work on versions of such problems where observations and action effects are modelled with probabilities, and the agent is acting to optimise some cost or reward function. This includes work on reinforcement learning and decision theoretic planning. To use rather more technical jargon we work on learning in unkown Markov Decision Processes (MDPs), near optimal learning in MDPs, reinforcement learning and modelling of unknown POMDPs (Partially Observable MDPs), policy search in POMDPs, planning and learning in structured MDPs (also known as factored MDPs, or Dynamic Bayes Networks).