Researchers are developing mobile robots that learn from their experiences to operate intelligently and independently. It is based on the robots developing an understanding of an indoor 3D space, such as an office environment, and how that space changes over time.
The four-year project known as STRANDS, or Spatio-Temporal Representations and Activities for Cognitive Control in Long-term scenarios, is a collaborative project being led by the University of Birmingham and involving security company G4S Technology. The robots will be tested in a security environment run by G4S.
Dr Nick Hawes, senior lecturer in intelligent robotics at the University of Birmingham, will co-ordinate the work across the eight sites taking part in the project. The research includes partners in Austria, Sweden, Germany and the UK.
He told SecurityNewsDesk the big challenge is giving the robots an understanding of how the world changes on a day-to-day basis and allowing them to detect situations that would be deemed potential security breaches.
The researchers will develop software to process huge amounts of data that the robots will encounter, allowing them to learn about their environment and recognise patterns of behaviour as the space changes over time. Experts from the University of Lincoln’s School of Computer Science will investigate creating 4D environment mapping and methods of detecting change and unusual situation.
Through the monitoring of recurring 3D shapes, objects, people and models of activity, the robots can monitor adaptive behaviour that, over time, will result in them able to tell the difference between a normal situation and an abnormal one.
Hawes claims the short-term security targets involve mobile observation, giving the robot a series of places to observe, defining times you want it to observe them and triggering it to contact a human guard if it observes something deemed ‘of interest’.
The researchers hope someday to be able to give the robots precisely defined rules for what is and isn’t allowed in the building.
“To patrol around it and send camera feeds back of particular areas and spot things you have told them to look out for, then that is quite within reach. We are working on systems at the moment that would be able to do that within a year or two in a research environment. And by the end of the project it would be ready for commercialisation,” he said.
“The common sense understanding of ‘what is a normal behaviour’ including people and how they move and what they do within the space, that is a much longer term project.”
All of the research is being devised to support human security officers, rather than replace them. The expert claimed that no matter how advanced they get with the technology there will always be a need for a “human level of filtering” after detection. There also remains a risk of robots missing things while they are learning the environment and issues with responding to detection, as the robot isn’t capable of making decisions.
Hawes said: “Ideally we want to equip the robot with enough knowledge to recognise things from day one, but there always needs to be humans in the loop.
“Whilst scientists would love robots to be fully autonomous and do everything for themselves, people that actually have to legislate and manage them are looking for a balance between robot autonomy and human control.
“What you want to have is enabling one human to control more and more robots, to have the same number of human guards providing a better quality of service by having robots to add to their ability to monitor the environment.”
The timeline for the project involves robots performing more advanced tasks for longer periods each year throughout the four-year period. There are developing targets for robot run-time and levels of difficulty with regards tasks they undertake. Run-times are an essential area for the project, as a short run-time means the robot is not able to gather enough information to learn a convincing statistical model of the environment. It is hoped that the robots will be able to run for around 120 days by the end of the project, dwarfing current state-of-the-art robots which are restricted to just a few hours of running.
Hawes claims the plan is to keep upgrading the artificial intelligence software to get the robots as smart as possible, while maintaining the same operating platform. He hopes the robot security guards will be ready for commercialisation at the end of the project and the researchers aim to engage with companies to develop appropriate platform hardware for varying situations.