It is perhaps understandable that the first robots to enter workspaces have been in quite neat and highly regulated environments. Recent evidence suggests, however, that robots are increasingly capable of functioning effectively, and safely, in more unstructured and cluttered environs.

A recent study from KTH Royal Institute of Technology highlights the progress that’s been made. The researchers developed a robot, called Rosie, that was capable of autonomously learning its way around a dynamically changing human environment.

Making Itself at Home

Rosie has lived in the Robotics, Perception and Learning lab at KTH for a few years now, and exists as part of STRANDS, which is a project to develop learning capabilities in robots, especially around their ability to perceive 3D environments and interact safely with them. The holy grail for this kind of work is to develop machines capable of operating in dynamic human environments, with people bustling around and the environment constantly changing.

Rosie maps the rooms at the lab and constantly revisits them to gain an updated picture. Its database typically has billions of points of physical space in it at any one time. The software is capable of modeling the static part of the environment, whilst simultaneously extracting the more dynamic aspects. It combines these to create a view plan around each dynamic element and gathers additional views to help it learn more.

It’s certainly an interesting project in what is a domain of growing importance. Check out the video below to see Rosie in action.