Dynamically adaptive human swarm interaction
Abstract
Advancement in swarm robotics has made it possible to develop systems that are adaptable and scalable. A practical application of this is in search and rescue missions where instead of using a single UAV operated by a single pilot to provide aerial coverage of the mission space, it is possible to deploy a swarm of UAVs controlled by a single swarm operator collecting data at n-times the rate with an increased efficiency in the human-to-robot mapping ratio. One of the major challenges associated with a single swarm operator controlling a swarm of UAVs is the interaction interface. What is the best way to present the status of each UAV in the swarm as well as the data collected by each agent to the swarm operator in real time without resulting in an information overload? One approach is to abstract the information in order to reduce workload. However, this could affect the explainability of the swarm actions and hence the operator's trust in the system.
The Human And Robot Interactive Swarm (HARIS) simulator is a web-based multi-agent simulator with human interaction at the core. It is specifically designed to test and optimise human swarm interaction and supports on-the-fly customisation of every aspect of the swarm's command and control. Heatmaps for the whole swarm and individual icons for each agent have each been found to be effective depending on the swarm size, number of tasks, and the user's cognitive workload. Certain scenarios such as time pressure and diagnosing errors may be better suited to a particular visualisation or abstraction approach. Therefore, we present a system that monitors the operator's cognitive workload and dynamically updates the interface to match their performance level. In this way, the user could benefit from the best available user interface configuration, and the system can help control their workload.
Our showcase demo will allow participants to interact with our simulator to perform a human swarm teaming task and experience the interface dynamically adapting to their change in workload due to task density increasing or decreasing while maintaining a high task completion efficiency. Participants will be able to operate the swarm and see the simulator abstracting individual drones into heatmaps and eliminating unnecessary data when they become too overloaded.
