• layered costmaps
• social navigation
• social robotics

The aim of this thesis is to implement a personalized autonomous robotic framework for providing useful services to humans in environments, such as factory or home places.
With the booming of the Industry 4.0 concept, “safety” is not the only main principle a robot must satisfy. Robots and people must share the same physical spaces and follow the same social behavioral rules. Therefore accomplishing a task with the expense of human comfort is not acceptable anymore. The robot has to perform motion actions and must be able to determine how to place itself relatively to a human and how to approach him in a relatively constrained environment by taking into account both safety, comfort and reliability; in a word the robot should be dependable.
For this reason, we present a Human-Aware Navigation system which takes into account safety, preferences and states of all the humans as well as the environment obstacles and generates paths that are not only collision free but also comfortable, in accordance with social conventions taken directly from the literature. The multi-sensor robotic platform developed, is able to identify a person compared to the surrounding environment and avoid him/her dynamically and in real time, adapting proxemics and speed to the person preferences, in order to maximize safety and comfort.
Rather than using conventional navigation approach, which tries to find the shortest collision-free path, not considering the semantical value of the obstacle, this thesis proposes a personalized path planning system based on the multiplane concept of layered costmaps. In layered costmaps the processing of costmap data is prosecuted on semantically-separated layers; each one tracks a type of obstacle or constraint, and then modifies a master costmap which is used for the path planning. This approach leads to a faster path planning and makes possible for robots to create complex navigation behaviors for different contexts and persons. Furthermore, this work tries to overcome and improve the actual concept of fixed proxemics widespread in literature, by creating adaptive layered virtual costs around humans, in order to produce different behaviors and paths compared to different people and their personal preferences, to the end of maximize comfort and safety.
The system has been developed on the Scitos G5 mobile platform equipped with laser range scanners and RGB-Depth cameras. Software has been implemented using the framework ROS (Robot Operating System), in order to enhance modularity and reusability. Data are taken from different sensors which are merged together to create a dynamic map used by the navigation.
We tested the system letting the robot interact with ten different people moving along a hallway, using a commercial wearable inertial system for record acceleration variations. The aim was to compare the navigation behaviors on user walking speed between the conventional navigation approach and the layered adaptive costmaps system proposed. The data show that our developed solution allows a mobile robot to navigate in a dynamic environment avoiding collisions with obstacles and people and, at the same time, minimizing discomfort in people by respecting spaces mentioned above.