Flag-based Human Supervised Robot Guidance Using Locality Preserving Projections

With the deployment of collaborative robotics in industrial and domestic applications, intelligent robots are anticipated to work with humans in dynamically challenging daily life environments. Robots must deal with several photometric and spatio-temporal complexity in the workspace to achieve successful human-robot collaboration. To overcome these issues, especially in the case of vision-sensor-based systems, several robustness and adaptability measures need to be incorporated. One of such could be the adaptation of a concise and distinctive feature extraction tool for information-rich, high-dimensional data such as locality preserving projections (LPP). The locality information of the original higher-dimensional data is encoded and preserved in that weight parameter unlike principal component analysis (PCA) and linear discriminant analysis (LDA). This can be very much useful in real-life vision-sensor-based systems where the intrinsic locality structure of the images carries significant information. This paper proposes an LPP-based solution for a real-life human-supervised robot navigation guidance system. The proposed approach outperforms contemporary feature extraction techniques like PCA.