Using direct competition to select for competent controllers in evolutionary robotics

Fitness functions in evolutionary robotics: A survey and analysis

“…Much of the ER research presented in the literature employs some form of hand-formulated, task-specific fitness function that more or less defines how to achieve the intended task or behavior. The most complex evolved behaviors to date consist of three or four coordinated fundamental sub-behaviors [14,[20][21][22]. In [14], the fitness evaluation method used was relatively selective for an a priori, known or predefined solution.…”

“…In [14], the fitness evaluation method used was relatively selective for an a priori, known or predefined solution. In [20][21][22] the fitness functions used for selection contained relatively little a priori knowledge, and allowed evolution to proceed in a relatively unbiased manner. This is an interesting contrast to much of the work aimed at evolving simple homing or object avoidance behaviors, many of which use complex fitness functions that heavily bias the evolved controllers toward an a priori known solution.…”

“…Here, the second robot might have received a higher fitness rating if it had not been obstructed by the first robot. Examples of the use of intra-population competition, in which the fitness of individual robots were directly affected by interaction with other robots using controllers from the same population, has been investigated in [21].…”

A behavior-based analysis of modal problems

“…In ER researchers start with a high-level description of a problem, such as developing a gait [2], autonomous navigation [20] or other complex tasks [19]. The strategy is to use evolution to search for robot behaviors that perform the desired task while introducing the smallest amount of prior knowledge into the search.…”

Exploring and evolving process-oriented control for real and virtual fire fighting robots

“…Nelson and Grant [11] investigate a complex Capture the Flag game, and evolve feedforward and recurrent neural networks in a competitive environment for this task in both virtual and real robots. Capi [3] also uses recurrent neural networks to learn a parallel foraging and protection task, and find successful controllers through the use of a multiobjective function.…”