Relevant for Research Area

C - Applications


With growing availability of robots and rapid advances in robot autonomy, their proximity to humans and interaction with them continuously increases. Humans in the vicinity of the robot should not only influence what the robot does but also how the actions are performed. In interaction scenarios with, e.g., autonomous cars or robotic assistants the policies of the robot should be adapted with regard to the preferences of the user. The goal of COBRA is to provide such brain state-informed robot adaptation. Based on decoding of brain signals from electroencephalography (EEG), the project partners will develop prediction methods for the perceived hazardousness and scene complexity in a human-robot environment. This information will be incorporated into the robot policy optimization in order to adapt the style of the robot’s action to its user. Ultimately, the aim is to provide a closed-loop system for human-compliant adaptation of robot policies based on the decoding of EEG signals. 

Research Status

The COBRA project started in March 2017 and will progress from passive scene observation in the first stage over decoding in robotic environments to online adaptation of robotic policies. Preliminary results [1] suggest that the hazardousness of events can be decoded from the user’s brain state in a driving scenario. When observing traffic scenes, responses not only differ between “surprising” events and normal driving, but also between hazardous and non-hazardous events.

[1] Kolkhorst H, Tangermann M, Burgard W (2017) Decoding Perceived Hazardousness from User’s Brain States to Shape Human-Robot Interaction. Proceedings of the Companion of the 2017 ACM/IEEE International Conference on Human-Robot Interaction, HRI ’17, pp. 349–350.