Online evolution in multirobot systems

ASSISI | bf blog post

In robotics, biological systems have provided an outstanding source of inspiration. In particular, Darwin’s theory of evolution has inspired researchers to study abstractions of natural evolution termed evolutionary algorithms to optimise the parameters of robotic controllers. A growing body of work has been devoted to online evolution [1], the idea of executing the evolutionary algorithm in the robots themselves while they perform their task. The goal is to provide robots with the capacity to learn new tasks and to respond to changes or unforeseen circumstances by modifying their behaviour in a completely autonomous manner. For example, consider a two-wheel robot with a gripper as shown in Figure 1. The robot’s task is to find and transport objects. If the gripper breaks during task execution, the robot can learn to push objects as a means to move them around.

Read the full blog post here