Jun Tani and Naohiro Fukumura. Learning goal-directed sensory-based navigation of a mobile robot. Neural Networks, 7(3):553-563, 1994.  Home/Search   Document Not in Database   Summary   Related Articles   Check

....training data from sensors and commands given, 2) obtaining an appropriate control function which fits the training data by means of supervised machine learning. Previously, back propagation neural networks have been used to train mobile robots to navigate environments via supervised learning [1, 2, 3, 4] however, these methods result in only limited success due to the following reasons: The architecture of the network is difficult to decide. Long, off line training times are often required. Uneven distributions of training exemplars can result in some patterns being over learnt and some ....

....exe mplars so that additional training patterns applicable to locations near the demonstrated paths are also obtained. Although just using the current state of the environment is unlikely to provide the robot with the ability to learn long or complex paths through environments (as was achieved in [1]) it has been demonstrated to be adequate for robots to perform simple reactive behaviours like wall following [10] and obstacle avoidance [12] In Section 2 of this paper we describe the fuzzy rule extraction algorithm used to generate the rule base as well as the means by which the training ....

[Article contains additional citation context not shown here]

J. Tani and N. Fukmura . Learning goal directed sensory-based navigation of a mobile robot. Neural networks, vol.7, No.3, pages 553-563, 1994.

.... qualitative map approach [Kuipers and Byun, 1991; Levitt and Layto, 1990] and the adaptive topological models introduced by [Prescott, 1994] Kurz s ultrasonic clustering techniques [Kurz, 1993] and the approach based on typical sequences of local sensorreadings rather than on explicit topology [Tani and Fukumura, 1994] as well as robust world modelling technique by Zimmer [Zimmer, 1996] are other qualitative approaches to self localization and navigation. The above qualitative navigation approaches concern spatial learning and path planning in the absence of a single global coordinate system for describing ....

J Tani and N Fukumura. Learning goal-directed sensory-based navigation of a mobile robot. Neural Networks, 7(3):553--563, 1994.

....the recognition of locations. The basic idea is to uniquely identify places by adding context information to their perceptual signatures. Nehmzow et al. [4] combine past and present sensorimotor experience of the robot in the input to a self organising feature map. Similarly, Tani and Fukumara [7] add previous sensor patterns to the input field of a neural network controller. A disadvantage of this approach is that the identical set of previously stored locations must be revisited Figure 1: The Manchester FortyTwo . The robot s sonar and infrared sensors are mounted on the turret. In ....

J. Tani and N. Fukumara, "Learning Goal--Directed Sensory--Based Navigation of a Mobile Robot", Neural Networks, Vol. 7, No. 3, pp. 553-563, 1994.

....3.2 Kalman Filter Formulation Unlike [2] in which we use heuristic to estimate the true depth of each region from multiple readings taken at each perceptual cycle, here a cascade of two filters (Fig. 3) and a sliding window of size three, to hold the present and the past two measurement profiles [17], are used. The first filter is a non linear median filter that estimates the current measured depth of a region j using, Z j;t = median i S j 1;t ; S j 2;t ; Z j N j ;t j (1) where S j i;t ; i = 1 : N j is the reading at time t of sensor i located in region j, N j the ....

Jun Tani and Naohiro Fukumura. Learning goal-directed sensory-based navigation of a mobile robot. Neural Networks, 7(31):553--563, 1994.

....a path from a given source to a given goal such that R can move along the path without colliding with obstacles. Many approaches to this problem have been proposed [3] Recently, neural network techniques have been applied in various ways to solve motion planning and related problems in robotics [1, 4, 5, 6, 7]. In particular, the Kohonen Self Organizing Map (SOM) 2] has been used to create a map of the free space of an environment [5, 7] Many known techniques require complete knowledge about position and shape of every obstacle and are therefore not applicable in situations where such prior ....

....motions are only used for backtracking) Hence, sensing the visible area in the forward direction is sufficient. In the simplified case of a point robot with unlimited motions, the interesting motions can be determined directly by considering local maxima in distances to the first obstacle [6]. With car like polygonal robots calculating the set of possible motions and extracting a small 1 This research was partially supported by the ESPRIT III BRA Project 6546 (PROMotion) and by the Dutch Organization for Scientific Research (N.W.O. robot Regular Mirrored 0.14 0.12 0.1 0.08 0.06 ....

[Article contains additional citation context not shown here]

J. Tani, N. Fukumura. Learning goal-directed sensory-based navigation of a mobile robot. Neural networks, Vol 7, No 3, 553-563, 1994.

....of sequences of plays, because such models are represented implicitly. It is also a problem that it is difficult to apply such neural networks to higher levels of decision making. However, these problems may be overcome by combining symbolic and neural planing using back propagation as feedback[7, 6]. The experiment described in this article was done as a first step of a study of learning cooperative behavior. Cooperation was not a major part in this experiment. We, however, established the experiment to be easy to extend cooperative learning, and we are planning further experiment in which ....

J. Tani and N. Fukumura. Learning goal-directed sensory-based navigation of a mobile robot. Neural Networks, 7(3):553--563, 1994.

....taking game theoretic approaches. Study for the behavior of society taking dynamical systems approaches are found in the new area called Ecology of Computation [22] or Emergent Computation [12] We have also started research in this direction, particularly on chaos and collective behavior [35][46][36] Distributed and massively parallel computing are expected to be powerful computing platforms [26] 7 Conclusions I have discussed a couple of things in this paper. Objects are things which can be distinguished from others. This notion brought us macroscopic programming . Concurrent objects ....

Tani, J. and Fukumura, N., Learning Goaldirected Sensory-based Navigation of a Mobile Robot, Neural Networks, in press.

....Their approach is to develop a representation in sensor space rather than any notion of real space. The system produces a representation, but the authors admit that the topological information generated is not useable for navigation purposes. A complete system based on SOMs was constructed by [Tani Fukumara, 1994]. They, too, built a sensor based representation of the environment and used that representation to build a relationship between sensor space and motor action. The relationship between the SOM and motor action was one to one, making this another SR model. It might be noted that a similar effect ....

Tani, J. & Fukumura, N. (1994). Learning goal-directed sensory-based navigation of a mobile robot. Neural Networks, 7, No 3, pp. 553-563.

....19] The second problem has been addressed by endowing robots with the ability to autonomously learn behaviors either from experimentation and dynamic interaction with their environment or via teleoperation. A number of learning algorithms have been used for this purpose including neural networks [29, 30, 37], evolutionary algorithms genetic programming [3, 6, 18] reinforcement learning [2, 13, 20] hill climbing routines [10, 27] and self organizing maps [16, 24, 35] In the context of robot navigation, a popular approach has been the construction and use of global maps of the environment [9] Such ....

Jun Tani and Naohiro Fukumura. Learning goal-directed sensory-based navigation of a mobile robot. Neural Networks, 7(3):553--563, 1994.

....The first type is skill based learning. In this approach the robot learns skills for achieving a fixed goal such as a homing or cyclic routing task. The robot has to go home or move into a predetermined cyclic loop, starting from an arbitrary position in the adopted workspace. Our previous work [42, 43] showed that the robot can achieve these tasks by acquiring an adequate state action map (a map of sensory based internal states to motor commands) The second type is model based learning, which is the main subject of this paper. The advantage of model based learning is that the process of ....

....steps can accumulate to a substantially large value in the middle of training which hampers the smooth convergence of the learning process. Numerous studies have been conducted of the problem of learning the sequential behavior of agents [6, 30, 47] including our prior work on skill based learning [43, 42]. These studies have shown that certain temporal internal representation are indispensable to the solution. Model based learning, presented here, differentiates itself in that its learning comprises not just learning sequences but also extracting grammatical structure hidden in the sequences. ....

J. Tani and N. Fukumura. Learning goal-directed sensory-based navigation of a mobile robot. Neural Networks, Vol. 7, No. 3, pp. 553--563, 1994.

No context found.

Jun Tani and Naohiro Fukumura. Learning goal-directed sensory-based navigation of a mobile robot. Neural Networks, 7(3):553-563, 1994.

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