Research Areas

Robots in swarms must have a sense of situational awareness to move in coordination with the other robots. Coordination among the swarm members in environments without absolute position providers such as GPS and motion capture systems requires the implementation of formation control algorithms which rely on inter-robot configurations such as relative positions. A robust realization of such coordination algorithms is even more challenging in the case of aerial robots because their decision layers usually demand generating estimation and control signals at relatively high data rates. We design and implement distributed onboard localization algorithms for robots to enable coordinated and collaborative behaviors in robot swarms. We evaluate quantitatively in simulations and real life experiments the performance of several combinations of the commercially available sensor suites including cameras, RGBD sensors, LIDARs, and ultra wideband range sensors. Furthermore, we investigate the effects of heterogeneity among swarm members in terms of sensing and motion capabilities.

In a variety of real life applications such as aerial coverage, border patrolling, and surveillance, a multi-robot system can perform superior to a single robot. However, designing resilient path planning algorithms for multi-robot systems remains an open problem. A typical path planning algorithm design consists of task allocation, inter-robot communication, and low-level trajectory tracking control. Essentially, the distributed coordination in an autonomous robot team can be achieved by deploying each individual robot in such a way that if every robot executes its own task, the entire team exhibits a coordinated behavior. We are interested in challenging scenarios where the robots are equipped with minimal sensing capabilities. For instance, in an adversarial environment, a drone team must communicate with short-range radio modules which allow limited situational awareness only. Our aim is to design coordination algorithms for robot swarms operating in uncertain, adversarial environments. Particularly, we investigate the game theory based methods which lead to real-time implementable distributed algorithms.

Recent years have witnessed tremendous developments in the field of artificial intelligence. As the performance of the commercially available computational boards have improved, implementation of the deep learning architectures onboard the ground and aerial robots in real time have become possible. Nowadays, several convolutional neural networks (CNN) architectures can be run on the small-size, low-weight boards such as Intel Jetson at reasonably high speed. In lab, we deploy the recently developed CNN and reinforcement learning architectures to solve the essential mobile robotics problems such as localization, mapping, and path planning. Remarkably, we focus on distributed frameworks for swarms that can be realized online.