Makes the Robot more Efficient
To have standalone machines, one solution is to build in an infantile state and give them the responsibility to learn to move and form. These tests are different sequences of movements.
For efficient robots, it is better to give them time to develop themselves their own abilities. According to this, like humans, robots evolving eventually learn to move faster than those designed in fixed form. But also to deal with an unexpected situation without losing their capacity. To confirm this theory, the project manager has developed virtual robots capable of evolving in a 3D environment. Each of these prototypes included a software that can test different movements until it gets to move first awkwardly then better, to a light source. The whole, depending on its structure, specific to each.
From 3D to reality
Each machine changed the shape. This is because it developed the ability to understand what part of the infrastructure it would need to move more skillfully. Another thing: each “generation” robot, time spent at the stage called “child” was becoming shorter. Once these simulations – the number of 5000 – the scientist has designed its equivalent with a set of lego. As virtual prototypes, it also learned to better exploit the entire structure to move more skillfully.
Why work on such a project? Because, says the researcher, robotic machines are currently scheduled to perform complex tasks such as paving a road. The various components that constitute – sensors, motors … – are programmed for it. But to take more simple but in unstructured environments and changing still more complicated to achieve. Particularly because it is assumed that the robot has adaptability. Hence the aim of Josh Bongard, developing robots able to develop their abilities rather than program behavior directly. The project is part of a more comprehensive large-scale production of automated machines, funded by the National Science Foundation.