Beyond black-box optimization: a review of selective pressures for evolutionary robotics

Abstract

Evolutionary robotics is often viewed as the application of a family of black-box optimization algorithms -evolutionary algorithms -to the design of robots, or parts of robots. When considering evolutionary robotics as black-box optimization, the selective pressure is mainly driven by a user-defined, black-box fitness function, and a domain-independent selection procedure. However, most evolutionary robotics experiments face similar challenges in similar setups: the selective pressure, and, in particular, the fitness function, is not a pure user-defined black box. The present review shows that, because evolutionary robotics experiments share common features, selective pressures for evolutionary robotics are a subject of research on their own. The literature has been split into two categories: goal refiners, aimed at changing the definition of a good solution, and process helpers, designed to help the search process. Two subcategories are further considered: task-specific approaches, which require knowledge on how to solve the task and taskagnostic ones, which do not need it. Besides highlighting the diversity of the approaches and their respective goals, the present review shows that many task-agnostic process helpers have been proposed during the last years, thus bringing us closer to the goal of a fully automated robot behavior design process.

Embodied artificial life at an impasse can evolutionary robotics methods be scaled?

“…The key motivation for studying these and similar tasks is that the coevolving populations may generate an adaptive fitness landscape that extends the duration of selective pressure (the Red Queen effect). However, as noted in [1] and others, coevolving populations can also enter into cycles of repeating competitive behaviors. More complex examples include [30] in which neural controllers are evolved for a task that combines pursuit and foraging.…”

“…These may explicitly select for some specific features of a solution but leave others unspecified alongside a weighted aggregate success/fail component. As noted in [1], the integration of multiobjective optimization methodologies A. L. Nelson in proc. 2014 IEEE Symposium Series on Computational Intelligence (SSCI '14) into ER has allowed for a much more nuanced approach to the definition of tailored fitness functions.…”

“…Although recent advancements, including the use of multiobjective optimization approaches as well as task-independent diversity maintenance methods, are likely to extend the range of evolvable ER behaviors by perhaps as much as an order of magnitude (reviewed in [1]), there remain fundamental difficulties in combining evolutionary methods inspired by nature with goal-driven searches.…”

“…In ER, by contrast, a specific behavior is usually sought, and hence there is often a A. L. Nelson in proc. 2014 IEEE Symposium Series on Computational Intelligence (SSCI '14) definable goal at least at a high level [1,20]. But this raises another difficulty: even with a well-defined goal, formulating a goal-oriented fitness function capable of traversing a vast ultrahigh-dimensional search space is likely to be intractable for complex tasks without the aid of detailed domain knowledge [21].…”

“…Doncieux and Mouret [1] review recent advances in ER methods in a survey of selective pressures used in evolution. These include diversity maintenance methods [2,[26][27][28] and the use of multiobjective optimization methods [29,30].…”