Simulation of Randomly Generated Domestic Environments for Assistive Robotics

The need for solutions that ensure good quality of life in ageing populations have raised interest on assistance robots, which operate in domestic environments. Consequently, high level of comprehension of indoor conditions usually containing all sorts of obstacles, even dynamic, and the identification of particular objects or workspaces are often the main challenges for robot manipulation and navigation. However, realistic domestic indoors are hardly accessible or suitable for testing and it is very difficult for researchers to obtain physical models with the necessary characteristics. This article presents a novel approach for randomly generating simulation indoor domestic environments, with the ability to add more detail, customise rooms and easy to use. This enables its use in a wide range of applications such as creating domestic images datasets, object detection, instance segmentation and classification, robot navigation and manipulation. The proposed approach allows to compensate for the lack of multi-purpose domestic environments in simulation available online. The results obtained show a high percentage of successful of environment generation with a reduced execution time. Nevertheless, improvements can be made to reduce the high generation time and unsuccessful completions that high level of detail bring with it.