Embodied evolution of robot bodies and minds in dynamic environments
|has title::Embodied evolution of robot bodies and minds in dynamic environments|
|Master:||project within::Technical Artificial Intelligence|
|Student name:||student name::Nicola Mularoni|
|Second reader:||has second reader::Berend Weel|
One of the most challenging quest within Evolutionary Robotics is the evolution of the robot controllers for a certain task in an on-line context. The difficulty of on-line evolution lies in the fact that control needs to be learned while actually performing the task at hand.
Previous work has shown how it is possible for an On-Line, On-Board Evolutionary Algorithm (such as Hybrid EvAg) to cope with dynamic environments producing robust and flexible controllers which were able to face different circumstances, such as wade a river or cross a narrow bridge. The way robots could tackle the different environmental circumstances was by creating robot-organisms. However, one limitation of this work is the fixed way in which the world is changing.
With this research we set out to extend the previous work in a scenario where the environment is changing continuously. In this scenario the robots are given a foraging task where they need to gather energy trough harvesting “plants”. Plants can exist in different sizes and consequently have different nutritional values. Furthermore the bigger the plant is, the bigger the robot organism needed to harvest it.
In this scenario the temperature of the world will be varied, which affects both the maximum speed of the robots, depending on the size of the organism, and the distribution of plants of different sizes.
The main research question is: Will the robots be able to adapt to the environmental changes using embodied evolution under the paradigms “cold makes you bigger” and “hot makes you smaller”?