This research deals with the developmental analysis of robot controllers that are created using evolutionary robotics (ER) methods. ER uses artificial evolution to automatically design and synthesize intelligent robot controllers. An aggregate fitness function that injects relatively little a priori task knowledge into the evolving controllers is used. The course of development of robot controllers evolving to perform a competitive goal-locating task is analyzed. To sample the course of evolution, controllers are taken from progressively more advanced generations, and are tested in a novel environment. Developments and changes in the controllers’ abilities and competencies are identified and correlated with overall controller fitness. As the evolution progresses, the robots evolved more complex high-level behaviors that are not explicitly selected for by the fitness function.
Sensorimotor integration, as it is applied to the ER algorithms that are tested on the Evolutionary Robotic (EvBot) platforms, will lead to greater sensor robustness, emergent intelligence, and autonomous behavior, in a similar manner to that seen in the natural world. The research will be carried out on the new EvBot III, see Figure 1. This autonomous robot platform has been design to have greater synergy between the hardware and the corresponding modular control architecture. Through effective sensorimotor integration, task learning time will be reduced, and the learned controllers will be more robust. Also, by combining and correlating multiple sensors and motor states, the system will be increasingly immune to noise and corrupted or missing data from the integrated sensors. These new methods for sensorimotor integration are biologically inspired. Cockroaches, see Figure 2, are chosen as the model for this research. Cockroach behavior and physiology are well characterized, and the cockroach model will be an objective measure for testing EvBot III robustness and its ability to operate autonomously in real-world situations.
Lastly, a modular control environment will speed up new EvBot III sensorimotor element prototyping, evaluation, and portability. The modularity of this new sensorimotor technology will enable systems, like the EvBot III, to adapt to constantly changing environments by integrating the necessary modules that allow the system to adapt to the task. The final system must exhibit robust control and robust behavior, this is critical for real-world tasks.