A simulation result of self-organizing control agents

Donggil Lee made an animation of a simulation result based on a recent draft “Decentralized Design and Plug-and-Play Distributed Control for Linear Multi-Channel Systems” written by Taekyoo Kim, Donggil Lee, and Hyungbo Shim.

This animation represents the situation depicted in the following figure.

The orange load cannot move itself, but want to be moved to the goal by the help of mobile robots (control agents). The mobile robots can push or pull the load when they attach to the load (in other words, join the control network), but they do not have sensors. The movement of the load is represented by fourth order linear system (two for the position and the other two for the velocity). The load can measure the relative position to the target, and this information is delivered to the attached robots.

Each robot is randomly moving around the load, and when an agent happens to meet the load, it receives the relative position from the load with which the velocity of the load should be estimated. A distributed observer is self-organized within the agent. For this, it needs the information of the input to the load (force exerted by all attached agents). This information is “distributedly” computed by communication with its neighbors, which is one of novel contributions of the paper. Moreover, in order to compute the force of the agent exerted to the load, a (state) feedback gain should be self-organized without any global coordinator or global information. Following the recipe of the paper, each agent self-organizes their dynamic controller. When the battery of the agent becomes empty, they leave the load (i.e., the control network), which is indicated by green color in the animation.