Helping robots understand their work will lead to safer workplaces

Robots that work alongside humans will need to understand the context of the tasks they are undertaking

Robots understand
Just holding something isn’t enough; it should understand why it is holding something

Currently, robots in industry are basically dumb machines: they do whatever their controlling algorithms tell them to do and no more, with no understanding of why they are doing it. But in order to work alongside humans safely, robotic systems will have to be much smarter, according to a new study from the National Centre for Nuclear Robotics at the University of Birmingham.

A team led by Dr Valerio Ortenzi has looked at how robots grasp objects and the actions to which that grasping is linked, and as they explain in Nature Machine Intelligence, they have come to the conclusion that a drastic change in the philosophy of robotics may be needed to make the interaction between human and automated systems safe and effective.

Currently, a “successful” grasp is one where the robotic gripper holds an object securely without causing damage. However, this may in fact be a real-world failure if, for example, the gripper is obscuring a crucial barcode which means that the object cannot be tracked. To be fully successful, the system has to understand the consequences of holding the item in the wrong way.

“Imagine asking a robot to pass you a screwdriver in a workshop,” Ortenzi said. “Based on current conventions the best way for a robot to pick up the tool is by the handle. Unfortunately, that could mean that a hugely powerful machine then thrusts a potentially lethal blade towards you, at speed. Instead, the robot needs to know what the end goal is, i.e.,to pass the screwdriver safely to its human colleague, in order to rethink its actions.”

In another example, Ortenzi noted that if a robot was passing a glass of water to resident in a care home, the action is successful if no water is spilled over the recipient and the glass could subsequently be successfully taken from the robotic gripper.

Such criteria are currently not programmed into the control algorithms for robotic systems. “The traditional metrics used by researchers, over the past twenty years, to assess robotic manipulation, are not sufficient. In the most practical sense, robots need a new philosophy to get a grip,” Ortenzi said. The team’s paper calls for a new metric for robotic operations to be devised centred on the task which is actually being performed, rather than just on the various actions that make up the overall task with no reference to whether the ultimate goal of manipulating the object is reached or not.

Commenting on the paper, the director of the NCNR, Prof Rustam Stolkin, said: “National Centre for Nuclear Robotics is unique in working on practical problems with industry, while simultaneously generating the highest calibre of cutting-edge academic research – exemplified by this landmark paper.”