C. W. Reynolds. Evolution of corridor following behavior in a noisy world. In From Animals to Animats 3: Proceedings of the Third International Conference on Simulation of Adaptive Behavior, pages 402--410, 1994. MIT Press.  Home/Search   Document Details and Download   Summary   Related Articles   Check

....sensoring information as described above. Fig. 2. Human s input data in polar coordinates (z = 2) All input values are linearly normalized into [0, 1] before they are entered into the neural controller. The input s format in polar coordinates is based on Reynolds work in artificial critters [7]. For the experiments presented in this paper z = 2, as it stresses the minimal amount of information for a Human to successfully achieve the desired behavior (i.e. for z = 1 neural controllers are not able to emerge satisfactory obstacleavoidance strategies) 2) Architecture: There has been ....

C. W. Reynolds, "Evolution of corridor following behavior in a noisy world," in From Animals to Animats 3: Proceedings of the Third International Conference on Simulation of Adaptive Behavior (SAB-94), D. Cliff, P. Husbands, J.-A. Meyer, and S. W. Wilson, Eds. Cambridge, MA: MIT Press, 1994, pp. 402--410.

....B. 186, 199, 201, 203, 209, 256] Porter, Brian, 122, 412] Probert, Penelope, 80, 176] Proychev, T. Ph, 251] Pun, F. 30] Pyylampi, Tero, 10] Quinn, Roger D. 216] Rabelo, Luis, 40] Ram, Ashwin, 86] Ramstein, E. 213, 261] Rana, A. S. 154] Ravichandran, B. 13] Reynolds, C. W. [112] Reynolds, Craig W. 67] Rice, James P. 396] Richter, R. 439] Rodriguez, A. O. 23] Ronge, Andreas, 241] Ross, Steven J. 242] Roston, Gerald Paul, 87] Routen, Tom, 140] Rouvinen, A. 310] Rudas, I. 135] Rush, J. R. 68, 88] Rusu, Calin, 142] Rutman, Nathan, 263] Ryu, H. ....

.... 256, 257, 262, 273, 280, 296, 308, 317] control , 276] coordination, 183] design, 165] hydraulic, 310] intelligent, 193] inverse kinematics, 224] learning, 111, 119, 158] legged, 66] locating, 286] manipulator control, 94] manipulator design, 260] manipulators, 199, 324] mobile, [343, 345, 369, 430, 58, 333, 348, 375, 425, 64, 70, 76, 87, 99, 112, 120, 123, 124, 130, 134, 137, 139, 141, 149, 152, 153, 158, 160, 163, 164, 166, 168, 177, 181, 185, 188, 191, 194, 196, 198, 202, 204, 215, 217, 222, 230, 238, 249, 252, 263, 267, 271, 286, 291, 292, 295, 297, 300, 304, 314, 320] modeling, 155] morphology, 302] motion, 282] motion planning, 368, 424, 92, 115, 116, 156, 247, 274, 283, 311] multi, 356, 425] multi , 214] multi arm, 154] multiple, 245, 259] navigation, 393] palletizing, 174] path eplanning, 335] path planning, 327, 328, 402, 409, 426, 427, ....

[Article contains additional citation context not shown here]

C. W. Reynolds. Evolution of corridor following behavior in a noisy world. In ?, editor, Proceedings of the Third International Conference on Simulation of Adaptive Behavior, pages 402--410, Brighton, UK, 8.-12. August 1994. MIT Press 1994, Cambridge, MA, USA. y(CCA37120/96) ga94bReynolds.

....were unsuccessful [4a] he was later able to demonstrate the validity of this approach in producing robustness The solutions discovered by this process are simple and robust. It appears that noise in fitness testing discourages strategies that are brittle, opportunistic, or overly complicated. [4b] However, this conclusion is valid only if the premise that a robust program is more resistant to noise than a brittle program is true. The latter premise might not be true for all domains, although what could be generalized is the fact that weaknesses of brittle code can be exploited to promote ....

Reynolds, C. W. (1994b) "Evolution of Corridor Following Behavior in a Noisy World", in SAB-94

....that the artifact needs to model the environment in some manner. This modelling requires the use of sensors. To achieve optimal performance from the artifact, it must have access to a range of potential sensors as well as the ability to integrate them into the model in an optimal manner. Reynolds [82] shows how performance is improved by not only letting the model determine a control algorithm, but by letting it place its sensors in appropriate locations. When using optimization methods to maximize mathematical functions, it is imperative that the method used converges to the correct result. ....

....different operating conditions or by changing the operating conditions for each generation of artifacts. For an extra terrestrial, exploration vehicle, the terrain to be traversed can typically be modelled stochastically. Another possibility is to present noisy data to the artifact s sensors [82]. A vehicle design that results from using only a single instantiation of the terrain for evaluation is likely to perform poorly in the actual settings, since it will not be the same as the terrain that was used for testing. However, by generating a number of terrains for testing, there is a ....

Craig W. Reynolds. Evolution of Corridor Following Behavior in a Noisy World. To appear in Simulation of Adaptive Behavior, 1994.

....programs for robot control is [6] Due to being overly simplified, however, it has been criticized as being not complicated enough to control a real robot [3] Another example of using a GP approach to evolve control programs is given in a series of papers by Reynolds. In the final version [8], the author applied arithmetic operations, such as , Gamma, and the conditional operation iflte to calculate a single output value from sensor values and interpreted it as the steering direction. The robot is then assumed to move for a fixed forward distance in the steering direction. In ....

C. W. Reynolds. Evolution of Corridor Following Behavior in a Noisy World. In Proceedings of the Third International Conference on Simulation of Adaptive Behavior, p402-410, 1994.

....It shares many of its basic principles with genetic algorithms [Gol89] but can be applied to a more general problem domain. Developed by John Koza and first published in [Koz92] the genetic programming paradigm has been successfully applied to several problems with mostly successful results [Rey94b, AP93, HWSS95, Tac93] Starting with a language consisting of functions and terminals, an initial population is produced by randomly combining the elements of that language into a tree like hierarchical program structure. Functions form the internal nodes of this tree, while the terminals form ....

....it performs a certain task. Programs which are being evolved for moving a lawn mower around an artificial lawn (i.e. the lawnmower problem from [Koz92] will receive a higher fitness based on how much of a lawn the program mows . Programs being used to guide a robotic cart through a corridor [Rey94b] receive higher fitness for manoeuvering the cart successfully through the corridor than programs that crash the cart. The above two problems are examples of error driven fitness a program s fitness is judged in relation to a known optimal solution or desired goal. In the lawnmower problem, ....

Craig Reynolds. Evolution of corridor following behavior in a noisy world. In Simulation of Adaptive Behavior (SAB-94), 1994.

....of termination criteria, the GP result is the best fit behavior that appeared in any generation. In addition to this generational process, a steady state evolution can be applied as in the Steady State Genetic Algorithm (SSGA) 7] which has recently been applied in GP for behavior evolution [8, 9]. In the SSGP approach the concept of generations does not exist. Instead, on each iteration following creation of the initial population only two new offspring are produced; the offspring replace two of the worst individuals in the population. In our approach, parent behaviors are selected to ....

Reynolds, C.W. "Evolution of Corridor Following Behavior in a Noisy World." 3rd Intl. Conf. on Simulation of Adaptive Behavior. (August, 1994), 402--410.

....simple simulated moving vehicle to avoid collisions. He comments that these solutions are brittle, vulnerable to any slight changes or to noise. In further work where the fitness testing includes noise, he reports that the brittleness problem is overcome, and only compact robust solutions survive (Reynolds, 1994). Floreano and Mondada (1994) were able to run a GA on a real robot in real time, rather than a simulation. The GA set the weights and thresholds in a simple recurrent network where every sensory input was connected to both motor outputs. The task was to traverse a circular corridor while avoiding ....

Reynolds, C. 1994. Evolution of corridor following behavior in a noisy world. In D. Cliff, P.

....of robot subsumption behaviors for wall following and box pushing. Shortly after the publication of Koza s text, applications of genetic programming to control problems of the type we focus on here have appeared in the literature. The most notable relation to this work is that of Reynolds [22]. He has used GP to evolve corridor following behaviors for a simulated robot in the presence of noise. Similar work has been done by Fraser [23] in evolving multi agent emergent behaviors, and Handley [24] in mobile robot path planning. We have already mentioned the difference between our ....

Reynolds, C.W. "Evolution of corridor following behavior in a noisy world", From Animals to Animats 3: Third International Conference on Simulation of Adaptive Behavior, MIT Press, pp. 402-410, August 1994.

....up to Sigma5 . Noise is extremely important in simulations for the diversity of occuring events and for the degree of generalisation reachable by the learning system. Reynolds showed that a noisy fitness enviroment allows for the development of more robust controllers in genetic programming [27]. Figure 6: The Khepera Simulator 4 The Memory based GP Control Architecture The memory based control architecture consists of two separate processes. One process communicates with the robot and stores past events into memory. The other process is constantly trying to learn and to induce a ....

C.W. Reynolds (1994) Evolution of Corridor Following Behavior in a Noisy World, in: From Animals to Animats 3: Proceedings of the Third International Conference on Simulation of Adaptive Behavior (SAB94), D. Cliff, P. Husbands, J.-A. Meyer and S. Wilson (eds.), MIT Press, Cambridge, MA.

....likely to be incorporated into the final, developed neural network. A possible way of combating the negative consequences of an experimenter s arbitrary choice is to let some aspects of an animat s morphology evolve in parallel with its control architecture, an approach already explored in [35] [40], 41] 9] 11 Finally, at this stage of our work, it is hard to draw any conclusion about the efficiency of the evolutionary algorithm used here. We happened on these specific settings after numerous trials and errors, which aimed at preserving over generations the diversity of the fitness ....

C. W. Reynolds, "Evolution of corridor following behavior in a noisy world," in From Animals to Animats 3. Proceedings of the Third International Conference on Simulation of Adaptive Behavior (D. Cliff, P. Husbands, J.-A. Meyer, and S. W. Wilson, eds.), pp. 402--410, The MIT Press/Bradford Books, Cambridge, MA, 1994.

....a real robot. Such approaches may even foreseeably involve both the overall morphology and the control system of these robots. In particular, several experiments have already been carried out in which the morphology of the sensors co evolves with the control architecture, both in simulation [59] and in hardware applications [60] 7 Conclusions This paper has described six recent approaches that combine an evolutionary algorithm and a developmental process in order to automatically design a neural network controlling the behavior of an animat. Some of these approaches also involved a ....

C. W. Reynolds. Evolution of corridor following behavior in a noisy world, in D. Cliff, P. Husbands, J.-A. Meyer, and S. W. Wilson, editors, From Animals to Animats 3. Proceedings of the Third International Conference on Simulation of Adaptive Behavior, pages 402--410. The MIT Press/Bradford Books, Cambridge, MA, 1994.

....simple simulated moving vehicle to avoid collisions. He comments that these solutions are brittle, vulnerable to any slight changes or to noise. In further work where the fitness testing includes noise, he reports that the brittleness problem is overcome, and only compact robust solutions survive [Reynolds94a]. Floreano and Mondada [Floreano94] were able to run a GA on a real robot in real time, rather than a simulation. The GA set the weights and thresholds in a simple recurrent network where every sensory input was connected to both motor outputs. The task was to traverse a circular corridor while ....

C.W. Reynolds. Evolution of corridor following behavior in a noisy world. In D. Cliff, P. Husbands, J.-A. Meyer, and S. Wilson, editors, From Animals to Animats 3, Proc. of 3rd Intl. Conf. on Simulation of Adaptive Behavior, SAB'94. MIT Press/Bradford Books, 1994.

....population abilities. Such a possibility has been first proposed by Hillis [40] and further explored in [41] 42] 43] The present work also demonstrates that, among the different paradigms that have been used to evolve the control architecture of an animat e.g. Lisp functions [21] [44], logic trees [45] 46] classifier systems [47] recurrent artificial neural networks exhibit several specific and attractive features. Besides being universal dynamics approximators as already mentioned, it turns out that they are low level, non specific primitives that can be combined to ....

C. W. Reynolds, "Evolution of corridor following behavior in a noisy world," in From Animals to Animats 3. Proceedings of the Third International Conference on Simulation of Adaptive Behavior (D. Cliff, P. Husbands, J.-A. Meyer, and S. W. Wilson, eds.), pp. 402--410, The MIT Press/Bradford Books, Cambridge, MA, 1994.

.... select parents from pop(t) recombine and mutate parents to create pop(t 1) determine fitness of pop(t 1) t = t 1 until best individual is good enough Figure 1: General Evolutionary Computation Algorithm of robot control architecture should be evolved There are a number of options: high level code [40], machine code [37] parameter settings for a hand designed system [39] situationaction rules [7] and entire rule based strategies [15] Perhaps the most innovative direction, however, is the combination of evolutionary computation with artificial neural networks. Neural networks allow the ....

C. Reynolds. Evolution of corridor following behavior in a noisy world. In D. Cliff, P. Husbands, J-A. Meyer, and S. Wilson, editors, From Animals to Animats: Proceedings of the Third International Conference on Simulation of Adaptive Behavior, pages 402-- 410, Cambridge, MA, 1994. MIT Press.

....minimum size. However more effort, that is more fitness evaluations, were required to evolve the correct program than when program size was ignored. Kinnear found he could evolve more general solutions by adding a term inversely proportional to program length to its fitness. Reynolds [Rey94b] introduced noise into fitness tests used to evolve programs to steer a simulated 2D vehicle through connected corridors at various angles. He concluded that the addition of noise into fitness tests discouraged solutions that were brittle, opportunistic, or overly complicated, so that the ....

Craig W. Reynolds. Evolution of corridor following behavior in a noisy world. In SAB-94, 1994.

....its environment by using the function look for obstacle, which returns a measure of the distance to an obstacle in the direction specified by its argument. As terminals he used the set of constants 0, 0.01, 0.1, 0.5 and 2. Reynolds also experimented with noise to evolve more robust controllers [27, 29] and investigated the influence the representation has on the difficulty of the problem [28] Reynolds experimented with fixed sensors and with sensors that could be pointed dynamically. Using fixed sensors simplified the problem considerably. Nordin and Banzhaf [22, 23, 24] used a miniature ....

C. W. Reynolds. Evolution of corridor following behavior in a noisy world. In D. Cliff, P. Husbands, J.-A. Meyer, and S. W. Wilson, editors, From animals to animats 3: Proceedings of the Third International Conference on Simulation of Adaptive Behavior, Brighton, England, 1994, pages 402--410. The MIT Press, 1994.

....normally rather skeptical about simulations, acknowledged in [Brooks92] that these constraints would probably require the use of simulation, but stressing again the dangers of this approach. Pure simulation is indeed an approach where many results have been achieved (e.g. Koza92] Reynolds93] [Reynolds94]) but never tested and validated on a real device. Nolfi, Floreano, Miglino and Mondada [Nolfi94] illustrate with several examples various methodologies that can be used in evolutionary robotics, spanning from careful simulation to experiments carried out entirely on real robots. The evolved ....

C. W. Reynolds. Evolution of corridor following behavior in a noisy world. In D. Cliff, P. Husbands, J. Meyer, and S. W. Wilson, editors, From Animals to Animats III: Proceedings of the Third International Conference on Simulation of Adaptive Behavior, pages 402--410. MIT Press-Bradford Books, Cambridge, MA, 1994.

....find the most appropriate solutions to satisfy simultaneously several sometimes conflicting goals. Within this latter approach, a number of researchers have successfully employed an evolutionary procedure [7] 8] to develop the control system of simulated robots [9] 10] 11] 12] 6] [13]. The rich variety of structures This research has been carried out at the Microcomputing Laboratory (LAMI) Swiss Federal Institute of Technology at Lausanne (EPFL) Switzerland. E mail: floreano, mondada di.epfl.ch. WWW: http: lamiwww.epfl.ch w3lami team mondada Financial support has been ....

.... WWW: http: lamiwww.epfl.ch w3lami team mondada Financial support has been provided by the Swiss National Research Foundation (project PNR23) that have been put under evolution (feed forward neural networks [9] dynamic recurrent neurons [11] 14] classifier systems [6] and Lisp code [10] [13]) and the large number of evolved behaviors (locating food sources, wallfollowing, obstacle avoidance, chemotaxis and tropotaxis, corridor following, light orientation, box pushing, gait control, etc. have empirically demonstrated the power and generality of the evolutionary methodology. However, ....

[Article contains additional citation context not shown here]

C. W. Reynolds, "Evolution of corridor following behavior in a noisy world", in From Animals to Animats III: Proceedings of the Third International Conference on Simulation of Adaptive Behavior, D. Cliff, P. Husbands, J. Meyer, and S. W. Wilson, Eds. MIT Press-Bradford Books, Cambridge, MA, 1994.

....of termination criteria, the GP result is the best fit behavior that appeared in any generation. In addition to this generational process, a steady state evolution can be applied as in the Steady State Genetic Algorithm (SSGA) 21] which has recently been applied in GP for behavior evolution [22, 23]. In the SSGP approach the concept of generations does not exist. Instead, on each iteration following creation of the initial population only two new offspring are produced; the offspring replace two of the worst individuals in the population. In our approach, parent behaviors are selected to ....

Reynolds, C.W. "Evolution of Corridor Following Behavior in a Noisy World." 3rd Intl. Conf. on Simulation of Adaptive Behavior, pp. 402--410, August 1994.

....optimal player. In the same book, Koza first discusses coevolution in Genetic Programming (pages 429 437) in the context of a discrete strategy game. In the work reported here, the vehicle model, and the noise tolerant Steady State Genetic Programming system was taken from [Reynolds 1994a] and [Reynolds 1994c] The vehicle model draws heavily from [Braitenberg 1984] and is equivalent to the turtle of the LOGO programming language. Competitive fitness and coevolution were first explored in evolutionary computation in the context of the Iterated Prisoner s Dilemma in [Axelrod 1984] Axelrod 1989] ....

Reynolds, C. W. (1994) Evolution of Corridor Following Behavior in a Noisy World, to appear in From Animals to Animats 3: Proceedings of the Third International Conference on Simulation of Adaptive Behavior (SAB94), in press.

No context found.

C. W. Reynolds. Evolution of corridor following behavior in a noisy world. In From Animals to Animats 3: Proceedings of the Third International Conference on Simulation of Adaptive Behavior, pages 402--410, 1994. MIT Press.

No context found.

C. Reynolds, #Evolution of Corridor Following Behavior in a Noisy World", in Simulation of Adaptive Behavior, 1994.

No context found.

C. W. Reynolds. Evolution of corridor following behavior in a noisy world. In [SAB94].

No context found.

Reynolds, C. (1994b), Evolution of Corridor Following Behavior in a Noisy World, in D. Cliff, P. Husbands, J.-A. Meyer & S. W. Wilson, eds, `From Animals to Animats 3: Proceedings of the Third International Conference on Simulation of Adaptive Behavior (SAB94)', MIT Press Bradford Books, pp. 402--410.

Online articles have much greater impact   More about CiteSeer.IST   Add search form to your site   Submit documents   Feedback  

CiteSeer.IST - Copyright Penn State and NEC