Towards Distributed Learning to Support Situational Awareness for Robotic Team Augmented Humanitarian Disaster Response

Abstract

The use of robots to assist first responders in disaster response has seen increasing adoption as underlying technologies mature. They are inherently adept at real-time knowledge acquisition and are able to perform a myriad of pre-stabilization tasks within hazardous environments in lieu of jeopardizing human lives. Heterogeneous robots operating in teams can collaborate to facilitate first responders’ decision making by providing situational awareness (what is happening and when). However, the cognitive load placed on the human must be carefully managed; too much information and the human becomes overwhelmed, too little, and the human becomes over-reliant on the autonomy and hence complacent. Both lead to poor outcomes, with the potential loss of lives. Our work in progress’s approach to this shared autonomy problem is to apply distributed machine learning to identify behaviors of interest from temporal changes to LiDar or video images. The overarching goal is a conceptual framework where a heterogeneous team of robots (aerial, ground robots equipped with different sensing, and computational capabilities) may rapidly learn the pertinent aspects of an unfamiliar dynamic terrain, and accordingly, improve the decision making and projection capabilities of human first responders. Essentially, robots working together to collect and disseminate actionable knowledge in the expedited manner to save human lives.