Causal Effect Estimation of Robot Actions for Human Aware Navigation

An autonomous robot that understands how its own presence and interaction with people can cause changes in the shared space is able to navigate more efficiently and to complete its tasks with greater success. This causal awareness also leads to greater safety in interaction, reducing situations of physical conflict between the robot and the humans sharing the same environment. Traditional approaches to navigation treat people as dynamic obstacles, reacting to their presence without modelling their behaviour or anticipating the consequences. This limits the robot's ability to act proactively and to exploit its own interaction as a tool to modify the environment in its favour. In the context of a university corridor, the aim is to study the causal effect on the environment produced by the interaction of a TIAGo robot with the people present. The robot stops at a safe distance, evaluates whether the passage is congested and may emit a signal to request that people move away. The entire interaction is modelled as a DAG over boolean variables, capturing the initial and final positions of the people, the robot's action, the resulting space and the outcome of the task.