Zsolt Kira

At work performed at the Navy Center for Applied Research in Artificial Intelligence (NCARAI) , I have research several aspects of multi-robot systems using a resource protection domain containing multiple defenders protecting a centralized resource from intruders.

The most current line of research investigates how an element of human control can be added to a robotic swarm that uses physicomimetics to produce formations [2]. Robotic swarms are capable of performing tasks with robustness and flexibility using only local interactions between the robots.  Such a system can lead to emergent behavior that is often desirable, but difficult to predict.  These properties make controlling swarms by a human operator challenging, both from the top-down perspective of global swarm characteristics as well as influencing individual swarm members.  For this work we presented two methods for controlling these aspects of a swarm and integrated these methods into one user interface, and showed how they can be used to influence the swarm in a defense scenario. Videos are coming soon!

As part of the Micro Autonomous Systems and Technology (MAST) project, I worked on incorporating new capabilities into MissionLab to support the tasking of multiple microvehicles for 3D exploration and navigation. This includes 3D behavior-based waypoint navigation, obstacle avoidance, and teleoperation. Support for such 3D navigation as well as sensing (via SICK range finders) was added at all levels, all the way from the hardware abstraction layer (hserver), the robot executable, and the mlab console layer. All code is available online on the MAST webpage here.

In an earlier line of research at NCARAI, I extended an approach called Anytime Learning (later renamed to Continuous and Embedded Learning) to multi-robots systems [1]. This work allowed the team to adapt to changing situations such as individual robot failures. For example, the experiments showed that the team could switch from generalists to specialists depending on the number of members in the team. This work was published at IROS 2007.