Data Generation in Simulated Domestic Environments for Assistive Robots

The increasing focus on assistance robots for aging populations in domestic settings highlights the critical need for solutions that ensure a good quality of life. One of the main challenges faced by these robots is the ability to understand complex indoor environments, which often contain various obstacles, including dynamic ones, and to identify specific objects or workspaces for effective manipulation and navigation. However, access to realistic domestic environments for testing is limited, making it difficult for researchers to obtain physical models that accurately reflect these conditions. This article presents a solution for collecting data from domestic environments through simulations using a randomized environment generation algorithm, data acquisition, and classification. This approach aims to compensate for the lack of accessible and versatile domestic environments available online for testing and improving the capabilities of assistive domestic robots. The results obtained demonstrate a high success rate in creating simulated data that can effectively applied to real robots and a high level of confidence from a neural network retrained with synthetic data.