Evolving Symbolic Controllers

Abstract

Abstract. The idea of symbolic controllers tries to bridge the gap between the top-down manual design of the controller architecture, as advocated in Brooks' subsumption architecture, and the bottom-up designerfree approach that is now standard within the Evolutionary Robotics community. The designer provides a set of elementary behavior, and evolution is given the goal of assembling them to solve complex tasks. Two experiments are presented, demonstrating the efficiency and showing the recursiveness of this approach. In particular, the sensitivity with respect to the proposed elementary behaviors, and the robustness w.r.t. generalization of the resulting controllers are studied in detail.

Evolution of Voronoi based fuzzy recurrent controllers

“…The representation and the algorithms have been validated in two problems in the area of system identification and evolutionary robotics. Future work include experiments on a real mobile robot and the study of the impact of using the so-called Symbolic Controllers approach [6].…”

Evolution of Voronoi-Based Fuzzy Controllers

“…The representation and the algorithms have been validated on a mobile robot obstacle avoidance problem run on a simulator in a four and eight (not shown) dimensions input space, and also on the inverted cart pole system (not shown). Future work include experiments on a real mobile robot, and comparisons with more classical grid representations for Takagi-Sugeno fuzzy systems, and the impact of using the so-called Symbolic Controllers approach [3].…”

Evolutionary Robotics: Exploring New Horizons

“…• decomposing the problem into sub-problems, each of them being solved separately, either implemented manually or learned. The resulting behaviors can then be combined through an action-selection mechanism, that may itself eventually be tuned through evolution [34,54,99]; • reformulating the target objective into an incremental problem, where the problem is decomposed into possibly simpler fitness functions of gradually increasing difficulties, ultimately leading to what is reffered to as incremental evolution [36]; • reformulating the target objective into a set of fitnesses optimized independantly in a multi-objective context [73]. As opposed to the previous point, a multiobjective formulation of the problem makes it possible to avoid ranking subfitnesses difficulties, which is often a tricky issue; • using co-evolution to build a dynamically changing evaluation difficulty in competitive tasks [79,96]; • changing the evaluation during evolution to focus first on simpler problems and make the robot face progressively more difficult versions of the same task [4]; • likewise exploring solutions of increasing complexity with mechanisms protecting innovation to give new solutions a chance to prove their value [94]; • searching for novelty of behavior instead of efficiency [60,61].…”