Slide1: Human-dog interaction as inspiration for future assistance robots Shaun Lawson School of Computing, Napier University, Edinburgh UK (s.lawson@napier.ac.uk)


Seizures and their prediction ROBOT WORKSHOP, VIKTORIA INSTITUTE, JAN 2005: Seizures and their prediction ROBOT WORKSHOP, VIKTORIA INSTITUTE, JAN 2005 It is estimated that 1 in every 133, or 5.5 million, people in Europe, have epilepsy, of which a quarter can have seizures which cannot be controlled by any therapy. A system that could detect the impending onset of a seizure has the potential to create a significant improvement to the quality of life of many such people. Almost all research in seizure prediction to-date has concentrated on the off-line analysis of recorded electroencephalographic (EEG) signals. However, no definitive conclusions on the sensor system required, the modelling and manual classification of typical data, or the most appropriate algorithms for the machine processing of data have emerged.


Seizure Alert Dogs® ROBOT WORKSHOP, VIKTORIA INSTITUTE, JAN 2005: Seizure Alert Dogs® ROBOT WORKSHOP, VIKTORIA INSTITUTE, JAN 2005 There is mounting evidence to suggest that dogs can, very reliably, predict the onset of a seizure in their owners. This ability can be exhibited by dogs with no formal training, however most are trained by organisations such as the Support Dogs charity in Sheffield in the UK. It is currently unclear as to exactly how dogs recognise pre-seizure conditions, though work to date indicates that they rely primarily on visual cues such as facial expressions, postures and general behaviour as opposed, for example, to olfactory and auditory cues. Other physiological cues such as muscle tension, respiratory and perspiration might also be monitored by dogs using visual, olfactory or auditory senses.


Are they any good at it? ROBOT WORKSHOP, VIKTORIA INSTITUTE, JAN 2005: Are they any good at it? ROBOT WORKSHOP, VIKTORIA INSTITUTE, JAN 2005 Other researchers, both clinicians and computer scientists, are usually very sniffy about Seizure Alert Dogs. Support Dogs have reported that, once trained, every dog in a study group of six animals was able to predict every subsequent seizure in their owner. The advance warning period between dogs did vary from 45 to 10 minutes but, once training for an individual dog was complete, this warning time did not vary. Interestingly dogs have also been shown to reduce the frequency of seizures over time. Support Dog - Harvey


Our project ROBOT WORKSHOP, VIKTORIA INSTITUTE, JAN 2005: Our project ROBOT WORKSHOP, VIKTORIA INSTITUTE, JAN 2005 A small inter-disciplinary research consortium has been formed: the HCI Group at Napier, the Canine Behaviour Centre, at the School of Psychology, Queen's University Belfast, and the Support Dogs charity. We also have a clinical partner to advise on ethics etc. We are attempting to exploit the ability of trained Seizure Alert Dogs® to inform the development of software and systems that can detect seizures well in advance of any previously proposed artificial methods. A proposal for a 3 year study is ready for submission to the UK EPSRC AIBACS (Adaptive and Interactive Behaviour of Animal & Computational Systems) programme.


Objectives of the project ROBOT WORKSHOP, VIKTORIA INSTITUTE, JAN 2005: Objectives of the project ROBOT WORKSHOP, VIKTORIA INSTITUTE, JAN 2005 To determine the stimuli used by trained Seizure Alert Dogs® when seizure alerting. To develop visual and physiological sensor systems to monitor, as closely as is possible, the same behavioural and physiological signs as the dogs and to use dogs to label or classify the output of the sensor systems. To apply and develop soft computing systems for autonomously classifying sensor data, acquired under realistic conditions, as pre-seizure or normal, and to use dogs to evaluate the effectiveness of such systems. To improve Seizure Alert Dog® selection and training and to assess the potential of creating assistive technology for deployment in homes, or worn by a user, in the future.


Dog-human interaction ROBOT WORKSHOP, VIKTORIA INSTITUTE, JAN 2005: Dog-human interaction ROBOT WORKSHOP, VIKTORIA INSTITUTE, JAN 2005 As is evident in the lack of current knowledge about how dogs perform seizure alerting, the complexities of dog-human social cognition are not yet fully understood. Pioneering work is ongoing in this area by the likes of Wells, Mills et al in the UK and elsewhere by Miklósi and Pongrácz et al. Some observations of dog-human inter- action have been used by computer scientists to inform the construction of a number of, mostly entertainment-based, systems. The Synthetic Characters group at MIT have reported a number of simulations that exhibit both dog-dog and dog-human interactions whilst dog-like robotic toys, most famously the Sony AIBO, can now be easily purchased.


Embodiment of systems? ROBOT WORKSHOP, VIKTORIA INSTITUTE, JAN 2005: Embodiment of systems? ROBOT WORKSHOP, VIKTORIA INSTITUTE, JAN 2005 If successful, our project will make a contribution towards future Electronic Assistive Technology (EAT) systems which aim to determine the state of health of an individual. It is inevitable that future EAT systems will also exploit advances in technologies such as ubiquitous, ambient and wearable computing and we have proposed passive sensor based solutions as a first generation system. However the social acceptance of ‘invisible’ pervasive monitoring systems is predicted to be patchy at best. Perhaps more significant for our work is that dogs probably use interaction with our humans to predict seizures and also that interaction with dogs reduces seizures. So in future, will we see Aibo wearing the Support Dog vest?