The ITALK project aims to develop artificial embodied agents able to acquire complex behavioural, cognitive, and linguistic skills through individual and social learning. This will be achieved through experiments with the iCub humanoid robot to learn to handle and manipulate objects and tools autonomously, to cooperate and communicate with other robots and humans, and to adapt to changing internal, environmental, and social conditions.

The main theoretical hypothesis behind the project is that the parallel development of action, conceptualisation and social interaction permits the bootstrapping of language capabilities, which on their part enhance cognitive development. This is possible through the integration and transfer of knowledge and cognitive processes involved in sensorimotor learning and the construction of action categories, imitation and other forms of social learning, the acquisition of grounded conceptual representations and the development of the grammatical structure of language.

The project will lead to the development of

• New theoretical insights, models and scientific explanations of the integration of action, social and linguistic skills to bootstrap cognitive development
• New interdisciplinary sets of methods for analysing the interaction of language, action and cognition in humans and artificial cognitive agents,
• New cognitively-plausible engineering principles and approaches for the design of robots with behavioural, cognitive, social and linguistic skills
• Robotic experiments on object manipulation and language with the iCub robot

Overall, the project proposes visionary research that will provide a new standard in embodied cognitive science and will demonstrate the effectiveness of the method proposed by integrating interdisciplinary theoretical and experimental research on a single advanced robotic platform.

Contact: Angelo Cangelosi
Link: www.italkproject.org

The CONCEPT project is an EPSRC funded project at the University of Plymouth in which we study how humans can teach concepts to robots. There are strong indications that young children and adults rely heavily on language when acquiring knowledge and in the CONCEPT project we aim to recreate this aspect of human learning on a robot. The robot will engage with people and will tap into their willingness to teach and share information. However, human learning is slow (it typically takes newborn children three years to actively use 300 words and related concepts) and to speed up the concept acquisition process we will study how robots, once they have learned something, can swiftly exchange that knowledge with other robots over the internet.

We believe that robot learning relies heavily on the interaction between the human and the robot. In order to facilitate this interaction we are studying novel methods of giving the robot a personality.

Contact: Tony Belpaeme
Link: www.tech.plym.ac.uk/SoCCE/CONCEPT/

The project use evolutionary robotics for the design of the neural controller of the Rover Mars robot.

This work has the objective to investigate the possibility of using an alternative sensor system, based on infrared sensors, for future rovers capable of performing autonomous tasks in challenging planetary terrain environments.

The simulation model of the robot and of Mars terrain is based on a physics engine. The robot control system consists of an artificial neural network trained using evolutionary computation techniques. An adaptive threshold on the infrared sensors has been evolved together with the neural control system to allow the robot to adapt itself to many different environmental conditions. The properties of the behaviour obtained after the evolutionary process has been tested by measuring the performance of the rover under various terrain conditions. Simulations results show that the robot, at the end of the evolutionary process, is able to avoid rocks, holes and steep slopes based purely on the information provided by the infrared sensors.

Contact: Angelo Cangelosi

Aim: The overall aim is the development of robot skills required for the manipulation of fabrics in an unstructured environment, e.g. home or laundry.

This project focuses on sorting tasks, such as those conducted before placing cloth in a washing machine.
The objective is the development of artificial vision and manipulation algorithms enabling a small humanoid robot to conduct fabric sorting tasks.

Method: The project starts with the observation of humans performing a cloth sorting task. (Figure 1A and 1B).
The observations were analysed (figure 2A) and formed the basis for a computational model of human visual search and grasp (Fig 2B).
This model is now being implemented.

Findings: Of special interest are following facts:
- The eyes saccade from grasping point to grasping point. There is no evidence of overt scanning of the visual scene for the selection of the next grasping point (figure 2A)
- Saccades targets reflect covert visual search and strategic planning processes, constrained for instance by the availability of the right or left hand for the grasping task and the motion of the hand after grasping.

Contact: Peter Gibbons, Phil Culverhouse, Guido Bugmann
Link: www.tech.plym.ac.uk/soc/staff/guidbugm/FabricManipulation/fabric.html

With the ever-increasing numbers of robots in the real world, in particular in industrial environments,
it is increasing important to build teams of robots capable of high level co-operation in real-time situations.

The complexity involved in multi-robot autonomous systems requires some form of model situation to
experiment and develop the inherent technologies. Robot football is clearly an intelligent game that brings together many engineering disciplines into a coherant whole.
(vision, mechanics, electronics, robotics A.I etc….)

The idea of FIRA Robot Soccer originated in 1995, with the first two competitions
(Mirosot '96 & Mirosot'97), being held in KAIST, Korea.

At the University of Plymouth we are continually developing a robot football team that currently
partakes in competitions governed by FIRA.(Federation of International Robot soccer Association).
FIRA Cup competitions bring together skilled researchers and students from different disciplines to play
the game of robot soccer. There are many categories involving different size robots, pitches, and various
levels of robot autonomy, which compete in different soccer tournaments.

• Humanoid Robots (HUROSOT)
• Single Humanoid Robot (S-HUROSOT)
• Micro Robots (MIROSOT 5a side & 11a side)
• Nano Robots (NAROSOT)
• Single Nano Robot (S_NAROSOT)
• Khepera Robots (KHEPERASOT)
• Khepera Robot (S-KHEPERASOT)

The University of Plymouth has taken part in the 5-a-side MIROSOT league for many years. It is now focusing on humanoid robot league: Hurosot in which it represents England in International competitions.

Contact: Guido Bugmann
Project page: http://www.tech.plym.ac.uk/robofoot/

This project explores a still-untapped method of knowledge acquisition and learning by intelligent systems: the acquisition of knowledge from Natural Language (NL) instruction. This is very effective in human learning and will be essential for adapting future intelligent systems to the needs of naive users. The aim of the project is to investigate real-world Instruction Based-Learning (IBL) in a generic route instruction task. Users will engage in a dialogue with a mobile robot equipped with artificial vision, in order to teach it how to navigate a simplified maze-like environment. This experimental set-up will limit perceptual and control problems and also reduce the complexity of NL processing. The research will focus on the problem of how NL instructions can be used by an intelligent embodied agent to build a hierarchy of complex functions based on a limited set of low-level perceptual, motor and cognitive functions. We will investigate how the internal representations required for robot sensing and navigation can support a usable speech-based interface. Given the use of artificial vision and voice input, such a system can contribute to assisting visually impaired people and wheelchair users.

Partners: University of Plymouth, University of Edinburgh.
Contact: Guido Bugmann
Project page: http://www.tech.plym.ac.uk/soc/staff/guidbugm/ibl/index.html

Aim: The overall aim is the development of human-robot interfaces allowing the instruction of robots by untrained users, using communication methods natural to humans.

This project focuses on card game instructions, in a scenario where a user of a personal robot wishes to play a new card game with the robot, and needs to first explain the rules of the game. Game instructions are a good example of more general instructions to a personal robot, due to the range of instruction type they contain: sequences of actions to perform and rules to apply.

The objective is developing a robot-student able to understand the instruction from the humans teacher and integrate them in a way that supports a game playing behaviour.

Method: The project starts with recordings of a corpus of instructions between a human teacher and a human student.
Starting a robot development project with recording users is an approach termed "corpus-based robotics"

Contact: Guido Bugmann, Joerg Wolf
Link: www.tech.plym.ac.uk/soc/staff/guidbugm/mibl/index.html

The aim of our research is to investigate visual object recognition in experts and apply that knowledge to machine recognition. In particular we are interested in expert perception of natural objects and scene rather than 'novice' or normal perception. Expert perception is characterised by a period of training, which is required to ensure that perceptions meet the criteria for expert behaviour. Projects include expert plankton categorisation and cytological smear slide assessment.

The work also extends our knowledge of visual perception in general.

Since 1989 the Natural Object Categorisation group at Plymouth University have been developing machine vision systems to categorise marine plankton. The group have focussed on the difficult task of discriminating microplankton, as it is a good model for investigating top-down influences of expert judgements on bottom-up processes. It has been particularly revealing to explore the issues of recognition in a target group of objects where natural morphological variation within species causes experts difficulties. An operational machine (known as DiCANN) has been constructed which has been extensively tested in the laboratory with field-collected specimens of a wide range of plankton species, from fish larvae and mesozooplankton to the dinoflagellates of the microplankton. DiCANN employs multi-resolution processing, ‘what and where’ coarse channel analysis with support vector machine categorisation.

The HAB Buoy project concluded with the construction of four Harmful Algal Bloom monitoring systems that have been deployed to partner sites in Italy, Spain and Ireland. The systems possess digital microscopes and DiCANN recognition software. Using a precision pumped water system, they sample 375ml per hour and image to 1 micron resolution. The DiCANN software is capable of recognising specimens that are greater than 20 micron, both phytoplankton and zooplankton, for monitoring purposes.

Projects include expert plankton categorisation, motion analysis, texture processing and cytological smear slide assessment.

Research topics:

• Visual Natural Object Categorisation
• Motion segmentation
• Assessing visual performance of experts
• Machine vision (MTutor on-line learning system) (Problem solving and expert-novice differences)

Contact: Phil Culverhouse
Link: newlyn.cis.plym.ac.uk/cis/

Slothbot is an idea of Mike Phillips realized with the help of the Robot Club of the University of Plymouth.

It is probably the slowest big robot in the world.

Slothbot locates itself by interrogating wirelessly the data page of the Arch-OS system that informs on various aspects of the state of the Portland Square Building, including the position of a slow robot in atrium B.

Contact: Mike Phillips
Link: Robot Club