An event at IET – Speaker: Professor Yiannis Demiris, posted by Sinéad Nolan.
Working on the CHIRON project, it is always helpful to attend events to see what other people are up to in the assistive technology world. With that in mind, I went to see Professor Yiannis Demiris speak at the ‘Robots Helping People’ event at the Institute of Engineering and Technology last week (11th May 2016).
Is this the future?
While many people hear the word ‘robotics in care’ and jump to conclusions based around the potential for worst case scenarios often imagined in the media (Channel 4’s Humans springs to mind), the reality at this event seemed rather different. Instead of a dystopian hell which featured robots instead of human carers, it was based more around assistive technology or assistive robotics becoming a supplement for certain basic services a carer might provide.
“I started the lab in Imperial called personal robotics for a particular reason – I have the vision that robots do not belong in factories,’ said Demiris. “They are robots that try to help us in any way they can. The reason I call them personal robotics is because one of the key issues is that we need personalisation.”
As anyone who has undertaken a project knows, things don’t always turn out as we plan them. That is why user engagement absolutely needs to be prioritised before building anything.
In one of Demiris’ most recent projects they tried to build smart wheelchairs for children with disabilities equipped with sensors to help the user navigate a corridor. An interesting thing happened in these studies – what they found was that many got frustrated when the wheelchair was controlled externally.
One of the mistakes, Demiris says, is for the engineer to assume they know what the user wants outright – instead they need to listen to what the user wants, and think about how they build a better way to make this possible.
‘Eventually, we tried to change the behaviour of the wheelchair to match the skills of the kid that was actually driving. These wheelchairs wanted to be able to adapt to the changing profile of the user.”
Instead of controlling the user, the user controlled the wheelchair – and the adaptive system only stepped in when needed (for example, if the child was about to bump into something).
Similarly, some of the robots Demiris designed aimed to observe how you move and tried to modify the trajectories for itself. This manifested in a robotic pair of arms that could carefully and gently help someone put their jacket on (see video here).
From an engineering stance this meant adaptive systems applicable to lots of different domains, whether it was physical cognitive, emotional or artistic.
“They all have the same underlying philosophy of getting some data from the user, building a model and adapting the behaviour to this user.
“We want to build the same technology for every person, and then the application area changes but the underlying technology stays the same. An interactive learning cycle which keeps going constantly with interaction with the user. We don’t use to one type of algorithm, we use lots.”
One of the key issues Demiris pointed out, were the challenges in trying to build a robot that could assist people.
“One is personalisation’s – we have to explicitly model the users’ parameters so we have to be able to adjust the behaviour based on these generic internal model – and then I will change the behaviour. It’s not useful to have the same type of behaviour for everyone.
“Prediction is a key point in our work, there is no point when you are building assistive robotics using traditional AI techniques which rely on collecting all your data, doing some sort of classification, then acting.”
Instead, once Demiris and his team have a model, they try to predict what kind of assistance this particular person needs. They personalise the assistance for each individual user and constantly adapt the assistance to different users.
“If you have, for example, something that helps you walk and prevents you from falling down, it’s no use for you to fall down and then to realise, ‘Ah I should have helped!’ For us the key point is actually doing prediction and prediction means collecting all the data trying to run in forward in time, trying to see if there will be some needs later and acting before the person needs the help. “If you act when the person has already fallen down, it’s too late,” he added.
But limitations and challenges aside, Demiris is positive about the future of assistive technology in the care sector.
“The time where we can have a robot help us for a minute has passed, we are going to have robots among us for longer periods of time and they have to get to know us, to serve us better.”