Mondada, F. and Floreano, D. (1995). "Evolution of Neural Control Structures: Some Experiments on Mobile Robots." Robotics and Autonomous Systems, 16, (pp. 183-195).  Home/Search   Document Details and Download   Summary   Related Articles   Check

....at angles of around 45 degrees or narrow passages. 3. 1 The rst experiment: obstacle avoidance The simplest behavior an autonomous robot should master is obstacle avoidance, and there exist many solutions to this problem, from Braitenberg vehicles [6] to neural network controllers [30] 14] [22]. Thus, the rst goal 7 a) b) Figure 2: Two example environments used for incremental evolution of robot controllers. a. A simple one, and b. a more challenging one. is to evolve a network which allows the Khepera robot to move in a given environment as long as possible without hitting any ....

Mondada, F. and Floreano, D. (1995), Evolution of neural control structures: Some experiments on mobile robots, Robotics and Autonomous Systems, 16, 183-195.

.... strictly reactive systems (Brooks 1986) or fairly homogeneous, with each behavior having a generic learning capability (Pebody 1995, Albus 1997) Often it is monolithic, as in systems that employ reinforcement learning, genetic algorithms, or neural networks to supervise control (Kaelbling 1993, Mondada Floreano 1995, Whitehead 1992, Koza 1991, Sims 1994, Harvey 1995) Vertebrate brains have specialization both by organs and regions within organs for storing relevant state, whether that state is fairly transient computations of perceptual or motor information or longer term skills and information (Carlson ....

Mondada, F. & Floreano, D. (1995), `Evolution of neural control structures: Some experiments on mobile robots', Robotics and Autonomous Systems 16, 183--195.

....Kenkyusho Shoho, 425] Konstruktion, 260] Machine Learning, 167] Mech. Mach. Theory, 235] Mechatronics, 280] Nippon Kikai Gakkai Ronbunshu C Hen, 155, 175, 220, 234, 367, 393] Robotersysteme, 399] Robotica, 132, 307, 323] Robotica (UK) 245] Robotics and Autonomous Systems, [124, 139, 144, 169, 216, 231, 247, 274, 279] Telematics and Informatics, 74] Teleoperators and virtual environments, 255] Trans. Inst. Electr. Eng. Jpn. C (Japan) 291] Trans. Inst. Syst. Control Inf. Eng. Japan) 298, 319] Transactions of the Society of Instrument and Control Engineers (Japan) 84, 113] total 71 articles in 40 ....

....[27] Emmanuel, T. 177, 185] Enns, Russell, 168] 14 Genetic algorithms in robotics Erbudak, M. 37] Erkmen, A. M. 37] Espenschied, Kenneth S. 216] Fagg, Andrew H. 401] Falkenauer, Emanuel, 43, 44] Farritor, Shane, 263] Feng, S. 400] Fennel, Theron Randy, 16] Floreano, Dario, [110, 134, 181, 124, 264, 293] Fogarty, Terence C. 129, 180, 212] Fogel, David B. 349, 350, 351] Fogel, Lawrence J. 349, 350] Ford, Gary P. 45] Fraser, A. P. 68, 71, 88] Fu, Li Chen, 39] Fujimoto, Hideo, 21, 36] Fujimoto, S. 243] Fujimoto, Shinsaku, 155] Fukuda, T. 173, 191, 275, 294] Fukuda, Toshio, 69, ....

[Article contains additional citation context not shown here]

Francesco Mondada and Dario Floreano. Evolution of neural control structures: some experiments on mobile robots. Robotics and Autonomous Systems, 16(2-4):183--195, December 1995. ga95Mondada.

....parents from the population based on their fitness. 5. Apply search operators to the parents and generate offspring which form the next generation. Figure 6: A typical cycle of the evolution of architectures. Considerable research on evolving ANN architectures has been carried out in recent years [33, 42, 45, 150, 151, 152, 153, 154, 155, 156, 157, 158, 159, 160, 161, 162, 163, 164, 165, 166, 167, 168, 169, 170, 171, 172, 173, 174, 175, 176, 177, 178, 179, 180, 181, 182, 183, 184, 185, 186, 187, 188, 189, 190, 191, 192, 193, 194, 195, 196, 197, 149, 198, 199, 200, 201, 202, 203, 204, 205, 206, 207, 208, 209, 210, 211, 212, 138, 213, 214, 215, 216, 118, 130, 127, 217, 218, 219, 220, 221, 222, 223, 128, 224, 225]. Most of the research has concentrated on the evolution of ANN topological structures. Relatively little has been done on the evolution of node transfer functions, let al..one the simultaneous evolution of both topological structures and node transfer functions. In this paper, we will analyze the ....

F. Mondada and D. Floreano, "Evolution of neural control structures: some experiments on mobile robots," Robotics and Autonomous Systems, vol. 16, no. 2-4, pp. 183--195, 1995.

....[Lund 1996, Miglino 1995] then a less accurate model is required, but it can take significant time to do the on line training on the actual robot. If the task can be completed and the fitness can be accurately judged in minimal time, all of the training can be done on line [Husbands 1997, Mondada 1995]. This method precludes the need for any model of the robot, but the training takes num generations num individuals time tocomplete task to do the training. An increase in any of these results in a multiplicative increase in training time. All of these techniques require that we either put ....

Mondada, F. and Floreano, D. (1995). "Evolution of Neural Control Structures: Some Experiments on Mobile Robots." Robotics and Autonomous Systems, 16, (pp.

....for example, testing the robot for only 200 actions for only 1 epoch will reduce the time consumption of a single evolutionary process in the example given above from 17 days to less than 2.5 days. This is the approach taken by researchers who do the evolution process entirely on the real robots [7, 8, 25]. The approach has the disadvantage of not allowing many testings of each single control system (few actions and only 1 epoch) so the performance of a single control system will be heavily biased by the initial starting position of the robot. Because of the problems with on line evolution ....

F. Mondada and D. Floreano. Evolution of neural control structures: Some experiments on mobile robots. Robotics and Autonomous Systems, 16:183--195, 1995.

....recharging area and return to this place at a given time. This behaviour is based on an accurate evaluation of the battery residual time and on an internal representation of the environment. In fact some of the hidden nodes displayed activation levels that clearly mapped the environment geometry. [41] The second example is [29] in which a team from the University of Sussex evolved a neural network to control the movement of a camera head mounted on a gantry, whose motion is designed to mimic that of a wheeled robot. In this case both the internal network, and the morphology of the visual ....

F. Mondada and D. Floreano. Evolution of neural control structures: some experiments on mobile robots. Robotics and Autonomous Systems, 16:183-- 195, 1996.

....only the probability term is changed. Yet, a random draw is done in order to change weights whose confidence term is low. Such a mechanism brings the robot to behave as if it was testing hypotheses. It is interesting to draw a parallel between the PCR algorithm and Genetic Algorithms (GA) see [26, 43]) The way binary weights are switched can be compared to the GA mutation process which allows bits change in the DNA sequence of an individual. Yet, while GA test the behavior of populations whose DNA is fixed, the PCR allows a similar process during the life time of a single robot (kind of ....

F. Mondada and D. Floreano. Evolution of neural control structures: some experiments on mobile robots. Robotics and Autonomous System, 16(2-4):183--195, December 1995.

....ANN. A common one is to train an ANN via a large set of sample data. This is called a supervised learning. However, in the case of autonomous behaviors, it is rather difficult to employ strictly supervised learning algorithms because the correct system output is not always available or computable [6]. In general, only some simple signs which indicate the output effect are obtainable from the environment. Therefore, a suitable way to train an ANN for autonomous behaviors may be to let it learn through constant interaction with the environment. By using and analyzing environmental feedback, ....

F. Mondada and D. Floreano. Evolution of Neural Control Structures: some Experiments on Mobile Robots. Robotics and Autonomous Systems, Vol.16, pp.183-195, 1995.

....and the vertical edges of the rectangular target. 2.8 Evolution Entirely on Robots Work by D. Floreano and F. Mondada is one of the very first attempts of evolving controllers entirely on a physical robot in real time without any human intervention. In Floreano Mondada (1994) and Floreano Mondada (1996) the authors describe successful results of evolving navigation and obstacle avoidance behaviors on a Khepera robot. This robot has proven to be the most successful platform for physical evaluation of genetic approaches to controller evolution, due to its small size and portability, as well as the ....

....In the first set of experiments, the authors demonstrated a robust collisionfree navigation behavior, but had to add a bending penalty to the fitness function in order to prevent a particular efficient solution that kept the robot spinning in a small circle within an obstacle free area. In Mondada Floreano (1996) the authors describe the approach applied to evolving a homing behavior. A decaying battery was simulated and the robot evolved a recharging behavior in which it learned an internal representation of the environment and moved toward the light (i.e. the recharging station ) which boosted its ....

[Article contains additional citation context not shown here]

Mondada, F. & Floreano, D. (1996), `Evolution of Neural Control Structures: Some Experiments on Mobile Robots', Robotics and Autonomous Systems.

....importance of using physical mobile robots as opposed to computer simulations. To our knowledge, those who used physical mobile robots only experimented with miniature robots such as the Khepera or a gantry robot. Most work in evolutionary robotics was done using neural net control architectures [5, 11, 12, 9, 7, 19, 8, 18]. In contrast to this work, we wanted to see if genetic programming [13, 15] can be used to evolve a hierarchical control architecture for a simple reactive navigation task on a large physical mobile robot. If evolution is carried out on a large physical mobile robot safety measures have to be ....

.... Psi where n is the number of fitness cases and D = t T and R s = jr s j t . Best possible raw fitness of zero is achieved if the robot avoids obstacles while performing a balanced number of turns to the right and left. A similar fitness function was previously used by Floreano and Mondada [7, 19, 8]. The adjusted fitness to be maximized is calculated according to fitness adj = 1 1 fitness raw . This fitness measure tries to maximize survival time while penalizing unbalanced turning. 4 Experiments First we performed a number of experiments using computer simulations and an environment ....

F. Mondada and D. Floreano. Evolution of neural control structures: some experiments on mobile robots. Robotics and Autonomous Systems, 16:183--195, 1995.

.... The first concerns the development of appropriate recurrent connection strengths on the motor neurons which prevents the robot from getting stuck in situations when contralateral sensors are equally activated, therefore generating behaviours which are more efficient than a feed forward controller [29]. The second properties is the direction of motion. Although the robot is perfectly circular and the fitness function does not specify the direction of motion, the evolved controller always moves in the direction with higher sensor density which gives a better resolution of encountered ....

F. Mondada and D. Floreano. Evolution of neural control structures: some experiments on mobile robots. Robotics and Autonomous Systems, 16:183--195, 1995.

....controller with 256 Kbytes of RAM and 512 Kbytes ROM manages all the input output routines and can communicate via a serial port with a host computer. Khepera was attached via a serial port to a Sun SparcStation 2 by means of a lightweight aerial cable and specially designed rotating contacts (see [34] for more detailed descriptions of technical issues related to the experiments described in this paper) In this way we could exploit the power and disk size available in a workstation by letting high level control processes (genetic operators, neural network activation, variables recordings) run ....

F. Mondada and D. Floreano. Evolution of neural control structures: some experiments on mobile robots. Robotics and Autonomous Systems, 16:183-- 195, 1995.

No context found.

Mondada, F. and Floreano, D. (1995). "Evolution of Neural Control Structures: Some Experiments on Mobile Robots." Robotics and Autonomous Systems, 16, (pp. 183-195).

No context found.

F. Mondada and D. Floreano, "Evolution of neural control structures: Some experiments on mobile robots," Robot. Autonomous Syst., vol. 16, nos. 2--4, pp. 183--195, 1995.

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