In this paper, we studied the problem of finding decentralized controllers for large-scale heterogeneous robot swarms exhibiting segregative behaviors. As seen in nature, Segregative behaviors involve sorting a group of robots into groups based on their type. Our approach involves learning controllers that utilize local information at test time by imitating the policy of a centralized controller based on a differential potential concept at training time. We parameterized our policy using a time-varying aggregation graph neural network with multi-hop communication. This incorporates information not only from immediate neighbors but distant neighbors. We showed that our controller outperformed a local controller that considers only immediate neighbors and achieved similar performance to the centralized controller through varied experiments. In addition, we demonstrated the scalability of our method by exploring larger swarms and different groups.

Different approaches to Visual-Inertial odometry(VIO) has been presented in literature. However, few research works exploits the quadrotor translational and rotational dynamics and the known thrust and torque inputs. Additional information from the dynamics with known control inputs improves the state estimation. This paper is focused on the estimation of the quadrotor UAV 6 DOF pose by fusing the information from the dynamics of the quadrotor coupled with visual-inertial measurements using an Unscented Kalman Filter. The thrust and torque inputs drives the prediction model while the VIO system provides 6 DOF pose, velocity and unbiased angular velocity of the vehicle. The results shows that our approach improves the state estimates by about 2-5% in translation and 37% in rotation.

This paper presents an efficient robot navigation model in a multi-target domain amidst static and dynamic workspace obstacles. The problem is that of developing an optimal algorithm to minimize the total travel time of a robot as it visits all target points within its task domain amidst unknown workspace obstacles and finally return to its initial position. In solving this problem, a classical algorithm was first developed to compute the optimal number of paths to be travelled by the robot amidst the network of paths. The principle of shortest distance between robot and targets was used to compute the target point visitation order amidst workspace obstacles. A hybrid modelling scheme premised on geometrical considerations was applied to determine the length of obstacles encountered by the robot. Also, a dynamic model premised on the collapse of the radiation cone was proposed and used to estimate the velocity of a dynamic obstacle along the robot’s path.