When is the best time to do the laundry?
The advent of a new power system will turn the distribution and consumption of electricity upside down. This, and driverless cars were discussed at a conference on artificial intelligence and planning in Delft.
Two significant changes are in store for us: the advent of the smart grid, and driverless cars. Many of the studies presented at the ICAPS event, one of the world’s largest conferences on artificial intelligence and planning, had to do with these two developments.
‘This field is all about automated solutions to planning problems’, says mathematician Dr Mathijs de Weerdt of the department of Software and Computer Technology (EEMCS) and one of the conference organisers. ‘We use reinforcement learning algorithms for that.’
De Weerdt is engaged in the optimisation of the smart grid. At present it is still easy to match electricity supply to demand. With their gas-powered plants, large electricity companies can increase or decrease production accordingly. But if solar and wind energy soon account for a greater share of production, that production may not be so flexible. ‘In that case, it’s important to create flexibility on the demand side. By lowering the price of electricity when there is a lot of wind or sunshine, you can encourage
people to do their laundry, for instance, at times when a lot of electricity is being produced. This is a very simple form of optimisation. The reality is much more complicated. You also have to take account of industry. Industrial processes themselves are very complicated to optimise. Factories cannot adapt their operations to the price of electricity just like that.’ With reinforced learning algorithms, De Weerdt is trying to make the system as flexible as possible.
Co-organiser and mathematician Dr Matthijs Spaan does similar research, but with the focus on driverless cars. He is involved in the i-CAVE project, a collaborative effort that includes TU Delft and Eindhoven University of Technology and which is investigating the most efficient way possible for semi-autonomous vehicles to deliver parcels. ‘The cars are driverless in certain areas that are relatively uncomplicated and for which we have detailed maps. They can't handle complex locations, like the city centre of Delft, independently, so there a driver has to take over. We are looking into how we can optimise this system so that as many parcels can be delivered as possible without the intervention of a driver.
But autonomous driving involves more than just route optimisation. Sign recognition is crucial as well.
Professor of Intelligent Vehicles Dariu Gavrila (3mE) is doing research on this. Artificial intelligence is extremely good at individual tasks, such as recognising images. But in traffic, you have to anticipate the actions of other road users in situations that are continuously changing. AI is not so good at doing that yet.
‘The scientific challenges lie mainly in the complexities of urban traffic and in dealing with cyclists and pedestrians’, said Gavrila two years ago during his inaugural address as a professor. Technologies that learn automatically, with the help of big data, what road users look like and how they generally behave, could help to better assess traffic situations.