Pfeifer, R. and C. Scheier, "Sensory-motor coordination: the metaphor and beyond," Robotics and Autonomous Systems, Special Issue on "Practice and Future of Autonomous Agents," vol. 20, No. 2-4, pp. 157-178, 1997  Home/Search   Document Details and Download   Summary   Related Articles   Check

.... reaching UTH Metta et al. 24] visually guided manipulation UTH Metta and Fitzpatrick [25] eye arm coordination RA Stoica [26] indoor navigation MR AG Weng et al. 27] Value system invariant object recognition MR AG Krichmar and Edelman [28] category learning MR AG Pfeifer and Scheier [29] perceptual categorization MR AG Sporns et al. 30] neuromodulation MR AG Sporns and Alexander [31] Categorization sensorimotor categorization AVH Berthouze and Kuniyoshi [32] invariant object recognition MR AG Krichmar and Edelman [28] sensorimotor categorization MR AG Scheier and Lambrinos ....

....to previous work, the modeled value signal had two additional features: a) its prolonged effect on synaptic plasticity, and (b) the presence of time delays [28, p. 829] Another instantiation of a value system, whose output was used as a gating signal to modulate Hebbian learning, is described in [29, 33] (see Categorization) Sporns and Alexander [31] tested a computational model of a neuromodulatory system in an autonomous Neuromodulatory systems are instantiations value systems that find justification in neurobiology. Examples include the dopaminergic and the noradrenergic systems. ....

R. Pfeifer and C. Scheier. Sensory-motor coordination: The metaphor and beyond. Robotics and Autonomous Systems (Special Issue), 20:157--178, 1997.

....in order to achieve robustness to more complex environments. Note we refer to functions not systems or modules . Pfeifer and Scheier discuss how perceptual categorization may be achieved by the use of sensory motor coordination, without the need for internal representations of stimuli [58]. Adaptive Resonance Networks investigate how stable storage and assimilation of mem For example, consider an agent evolved to stand still, in the face of many perturbations, in a realistic physical simulation. The tness function being identical to the one employed here, except that 1000 ....

Rolf Pfeifer and Christian Scheier. Sensory-motor coordination : the metaphor and beyond. On-line paper., ??

....updated onto the SES where the position of an end e#ector would be automatically associated with other sensory events that occur in the same place at the same time. The association of sensing with action to form descriptions of sensory motor coordination (SMC) events has been shown by Pfiefer [20], Cohen [21] Grupen [22] and others [19] 23] to foster the self organization of information into categories for recognition. The SES facilitates SMC through the confluence of exteroceptic sensory streams with those related to body pose, joint velocities, forces, torques, etc. A coupling of the ....

Pfeifer, R. and C. Scheier, "Sensory-motor coordination: the metaphor and beyond," Robotics and Autonomous Systems, Special Issue on "Practice and Future of Autonomous Agents," vol. 20, no. 2-4, pp. 157-178, 1997

....which corresponds to an appropriate behaviour. Animals, however, exhibit a dynamically evolved See [22] for approximations in higher dimensions Visit [APPENDIX B:VI] for definition of the notation Refer to the following for various approaches to the design principles of autonomous agents: [18,19,21,24,25,26,31] The work in this section is influenced by the author s work at [38] interaction with the environment, which they affect and are affected by constantly. Given that the latter is not static, such an approach to the issue of behaviour selection is inadequate for creatures that have multiple ....

Pfeifer R., Scheier, C. "Sensory-Motor Coordination: The metaphor and Beyond ", Robotics and Autonomous Systems, special issue on `Practice and Future of Autonomous Agents', 1996

....updated onto the SES where the position of an end e#ector would be automatically associated with other sensory events that occur in the same place at the same time. The association of sensing with action to form descriptions of sensory motor coordination (SMC) events has been shown by Pfiefer [20], Cohen [21] Grupen [22] and others [19] 23] to foster the self organization of information into categories for recognition. The SES facilitates SMC through the confluence of exteroceptic sensory streams with those related to body pose, joint velocities, forces, torques, etc. A coupling of the ....

Pfeifer, R. and C. Scheier, "Sensory-motor coordination: the metaphor and beyond," Robotics and Autonomous Systems, Special Issue on "Practice and Future of Autonomous Agents," vol. 20, no. 2-4, pp. 157-178, 1997

....of the action. Recently, we have shown that a small number of repetitions of a humancontrolled robot task is sufficient to extract a sequence of SMC event descriptors for the task [29] The descriptors are equivalent to the categories that Pfeifer has shown to form when a robot learns SMC [30]. An SMC event descriptor represents the sensory conditions under which a motor action should be performed. In that respect it describes a basic behavior. If one casts the SMC event descriptor in terms of sensory preconditions and sensory post conditions, it is synonymous with a competency module ....

Pfeifer, R. and C. Scheier, "Sensory-motor coordination: the metaphor and beyond", Robotics and Autonomous Systems, Special Issue on Practice and Future of Autonomous Agents, vol. 20, no. 2-4, pp. 157-178, 1997.

....can take advantage of this ability in different ways. We will refer to the process of exploiting the agent environment interaction (i.e. the ability to select sensory patterns that are useful for some purpose through certain motor actions) as sensory motor coordination (for a similar view see [1]) We will show three different ways in which sensory motor coordination can help to solve otherwise insoluble problems. The way in which sensory motor coordination can be exploited also depends on the characteristics of the agent. As shown in [2] for example, agents that are able to modify ....

Pfeifer, R. & Scheier, C. Sensory-motor coordination: The metaphor and beyond. Robotics and Autonomous Systems. 20 (1997) 157-178.

....continuous learning with predefined abilities. Robotics studies tend often, when developing a learning model, to address a particular problem of sensor actuator coordination, e.g. maze travelling ( 30] 39] spatial navigation and exploration ( 13] 14] 25] or object manipulation ( 1] [34]) where often only one direction of control (from sensor to actuator) is considered. In contrast, our approach tries to develop a single control architecture which enables a robot to learn and act independently of a specific task, environment or robot used for the implementation. For this we look ....

Pfeifer, R. & Scheier, C. (1997), `Sensory-motor coordination: the metaphor and beyond'. In R. Pfeifer and R. Brooks (Eds), Robotics and Autonomous Systems, special issue on practice and future of autonomous agents." 20, Nos. 2-4.

.... a learning model, to address a particular problem of sensoractuator coordination, e.g. maze traveling (Owen Nehmzow, 1996; Tani et al. 1997) spatial navigation and exploration (Floreano Mondada, 1996; Gaussier et al. 1998; Kuipers, 1987) or object manipulation (Asada et al. 1997; Pfeifer Scheier, 1998), where often only one direction of control (from sensor to actuator) is considered. In contrast, our approach tries to develop a single control architecture which enables a robot to learn and act independently of a specific task, environment or robot used for the implementation. For this we look ....

Pfeifer, R. & Scheier, C. (1998). Sensory-motor coordination: The metaphor and beyond. Robotics and Autonomous Systems, Special Issue on Practice and Future of Autonomous Agents, 20, 2-4.

.... a unique expert robot [6, 9, 16] 2) whether the use of explicit communication could improve the performance of a group of robots in a collaborative task ( 1] 2] 8] 17] 22] 3) what learning abilities should the robot(s) be provided with for adapting to a continuously changing environment [7, 14, 18, 21]. We address these three issues in a specific task, namely learning the topography of an environment whose features, the locations of objects, change frequently. The locations of objects are learned by a group of worker robots which constantly search the environment, and which communicate to each ....

Pfeifer, R. and Scheier, C., (1998), `Sensory-motor coordination: the metaphor and beyond', R. Pfeifer, R. Brooks (eds.), Robotics and Autonomous Systems, special issue on practice and future of autonomous agents, 20:2-4.

....basic properties that have to be integrated in an arti cial vision system. First of all, one must notice the fundamental ambiguities of perceived informations. The action can lead to a new meaning of an object or can be used to suppress the ambiguity (moving around an object, turning the object (Pfeifer and Scheier, 1996)) The fundamental aspect of the interactions between perception and action has been emphasized by Gibson in its ecological approach of vision processes (Gibson, 1986) In the same vein, it would be nice if robots could be able to nd by themselves relevant visual information according to a ....

....neuron on the OR map. The action of the robot can be used to link the neurons associated with the di erent canonical views and build a graph for the 3D aspect of the object. With such a system, recognizing an object is equivalent to retrieve the same pattern of action (the robot moves around it (Pfeifer and Scheier, 1996)) Each image recognition and action can predict the next view. Thus, a visual transition is a movement and reciprocally. These considerations bring us back to navigation systems and planning theories. 5. Conclusion This work shows it is possible to extend a place recognition system to object ....

Pfeifer, R. and Scheier, C. (1996). Sensory-motor coordination: the metaphor and beyond. Robotics and Autonomous Systems.

.... a unique expert robot [5, 7, 13] 2) whether the use of explicit communication could improve the performance of a group of robots in a collaborative task ( 1] 2] 8] 14] 17] 3) what learning abilities should the robot(s) be provided with for adapting to a continuously changing environment [6, 11, 15, 16]. We address these three issues in a specific task, namely learning the topography of an environment whose features, the locations of objects, change frequently. A group of worker robots search constantly the environment. The robots are provided with an associative memory which allows them to ....

Pfeifer, R. and Scheier, C., (1998), `Sensory-motor coordination: the metaphor and beyond', R. Pfeifer, R. Brooks (eds.), Robotics and Autonomous Systems, special issue on practice and future of autonomous agents, 20:2-4.

....obstacle avoidance behaviors (Braittenberg, 1984) In order to allow more flexible behaviors it has been necessary introducing learning capabilities. At this point, conditioning paradigm has been used as inspiration to build learning rules (Verschure et al. 1995; Pfeifer and Verschure, 1994; Pfeifer and Scheier, 1996). Those two levels are similar to the two first steps of the generic sumsumption architecture proposed by Brooks (Brooks, 1981) The next step would be to learn how to plan. Yet, this approach seems confronted with the inverse problem of the one encountered by cognitive approach: it is very hard ....

Pfeifer, R. and Scheier, C. (1996). Sensory-motor coordination: the metaphor and beyond. this issue.

....of neurons must be added between the input and the output of the system. If unsupervised learning is to be used, that layer can be a WTA group (Winner Take All [26] an ART architecture (Adaptive Resonance Theory [6] 1 for an application) or a quickly learnable version of the Kohonen map ([29, 30] for applications) which, in essence, is a statistical classifier. The more examples of a class are presented, the more neurons are used to represent the class on the map.Therefore, if an insufficient number of instances of a given class is presented to the network, that class will be forgotten. ....

R. Pfeifer and C. Scheier. Sensory-motor coordination: the metaphor and beyond. this issue, 1996.

....further sensing. Hence, the machine must be a robot that can learn. That hypothesis moves the complete agent perspective from one of observation to one of creation. A fundamental idea in the development of intelligence in a situated, embodied agent is that of sensory motor coordination (SMC) In [26] Pfeifer and Scheier list five reasons for the importance of SMC in the construction of an intelligent robot. Sensory motor coordination 1. provides the basis for physical control over objects, 2. implies that both sensory and motor processes play an integral role in perception, 3. induces ....

....if it is not coupled with a capacity to learn those sensory to motor couplings that lead the robot to success. Learning SMC may be accomplished in a number of ways, all of which entail the forming of spatio temporal associations between sensory and motor events. Examples of this are given in [26]. Learning within a behavior based context that employs (or is applicable to) SMC has also been described by Maes [18] by Michaud and Mataric [19] and by Billard [3] among others. A consequence of behavior based design using SMC is that the robot is guided by an implicit value system rather ....

[Article contains additional citation context not shown here]

Pfeifer, R. and C. Scheier, "Sensory-motor coordination: the metaphor and beyond," Robotics and Autonomous Systems, Special Issue on "Practice and Future of Autonomous Agents," vol. 20, No. 2-4, pp. 157-178, 1997

.... unique expert robot [6, 10, 19] 2) whether the use of explicit communication could improve the performance of a group of robots in a collaborative task ( 1] 2] 9] 20] 25] 3) what learning abilities should the robot(s) be provided with for adapting to a continuously changing environment [8, 17, 21, 24]. We address these three issues in a specific task, namely learning the topography of an environment whose features, the locations of objects, change frequently. A group of worker robots search constantly the environment. The robots are provided with an associative memory which allows them to ....

Pfeifer, R. and Scheier, C., (1998), `Sensory-motor coordination: the metaphor and beyond', R. Pfeifer, R. Brooks (eds.), Robotics and Autonomous Systems, special issue on practice and future of autonomous agents, 20:2-4.

....architecture of one successful evolved individual (the architecture was subjected to the evolutionary process) they found that it was able to solve the task by relying only on the 5 This does not imply that evolutionary robotics is the only possible methodology for studying adaptive behavior. Pfeifer and Scheier (1997), for example, proposed a list of design principles which should help researchers both to design artificial agents and to understand natural organisms. Another possibility, of course, is to use evidence from neuro physiology when it is available at a sufficient level of detail (see, for example, ....

Pfeifer, R. & Scheier, C. (1997) Sensory-motor coordination: The metaphor and beyond. In R.

....internal representation through interaction between an agent and its environment. In the field of computer vision, purposive vision have proposed [1] 2] In simple terms, this approach suggests that vision has a purpose, and Perception Action system should be organized by behavior. Pfeifer [8] proposed the concept of sensorymotor coordination. It means that all interaction with the environment is to be conceived as a sensorymotor coordination. As a learning algorithm, reinforcement learning is often utilized. Reinforcement provides agents with the capability of learning to act ....

Rolf Pfeifer and Christian Scheier. Sensorymotor coordination: the metaphor and beyond. In R. Pfeifer and R.A. Brooks, editors, Robotics and Autonomous Systems. Special Issue on Practice and future of autonomous agents., 1996.

....must be added between the input and the output of the system. In the case of unsupervised learning, it can be a WTA group (Winner Take All [40] an ART architecture (Adaptive Resonance Theory [9] 3] for an application) or a quickly learnable version of the Kohonen map ( 27] 45] [47] for applications) which remains a statistical classifier: the most examples of a class are presented, the most neurons are used to represent the class on the map. Hence, if not enough examples of a class are presented, they are forgotten, which can have dramatic effects for an autonomous robot ....

.... Life (AL) approach, nowadays, most contributions on autonomous robots deal with simple universes in which evolve one or several simple robot(s) The collective tasks implemented are obviously very interesting because of their emergent behaviors: simple robots are able to generate complex behaviors [12, 38, 18, 47]. The adaptation capability of those robots does not lean on any internal adaptation mechanism but only supposes to take into account the feedback loop on the environment during the design process (constructivist approach [39, 62, 56] However, those robots require simple and explicit sensors ....

R. Pfeifer and C. Scheier. Sensory-motor coordination: the metaphor and beyond. this issue, 1996.

No context found.

Pfeifer, R., and Scheier, C. (1997). Sensory-motor coordination: the metaphor and beyond. Robotics and Autonomous Systems, 20, 157-178.

....the real world does not deliver neat feature vectors, but continuously varying sensory stimulation which, in addition, strongly depends on the agent s current behavior. If categorization worked indeed by mapping feature vectors onto category nodes (which is highly doubtful, cf. Edelman, 1987; Pfeifer and Scheier, 1997, 1999) this sensory stimulation would have to be translated into static feature vector, a highly non trivial task for which there is currently no general solution. While there is some research in adaptive behavior that deals with morphology (e.g. Cliff and Noble, 1997; Dellaert and Beer, 1996; ....

....morphology and neural processing is essential, is sensorymotor coordination. There is a lot of research demonstrating the importance of sensory motor coordination for perception, categorization, and concept development (e.g. Clancey, 1997; Dewey, 1896; Edelman, 1987; Kuniyoshi and Berthouze, 1998; Pfeifer and Scheier, 1997, 1999 [chapter 12] Thelen and Smith, 1994, to mention but a few) While sensory motor coordination serves the purpose of controlling objects in the real world, it has, in addition, information theoretic effects: through sensory motor coordination itself, patterns of sensory stimulation are in ....

Pfeifer, R., and Scheier, C. (1997). Sensory-motor coordination: the metaphor and beyond. Robotics and Autonomous Systems, 20, 157-178.

....easy to add control to make it go over obstacles. A picture of a passive dynamic walker is shown in figure 1. The passive dynamic walker is an example of cheap design , meaning that it exploits the physics and the system environment interaction which makes it cheap and parsimonious (see also [23, 25]) This approach is quite in contrast to the one taken by the Honda design team where the goal was to design a humanoid robot (i.e. a robot that looks like a human) that could move its limbs into every possible position and thus perform a large number of possible movements. By contrast, all the ....

.... in the real world is not a computational problem, or at least not an exclusively computational one and requires that embodiment be taken into account is gaining increasing acceptance: It has been demonstrated that categorization is best viewed as a process of sensory motor coordination [6, 23]. The term sensory motor coordination which goes back to John Dewey 1896 [5] designates processes where there is a coupling of sensory and motor processes with respect to a particular purpose. For example, a robot which is turning about its own axis is not involved in a sensory motor coordination ....

[Article contains additional citation context not shown here]

Pfeifer, R., and Scheier, C. (1997). Sensory-motor coordination: The metaphor andbeyond. Practice and future of autonomous agents [Special issue, R. Pfeifer and R. Brooks (Eds.)]. Robotics and Autonomous Systems, 20, 157--178.

No context found.

Pfeifer, R. and Scheier, C..: Sensory-motor coordination: the metaphor and beyond. In R. Pfeifer, and R. Brooks (eds.). Robotics and Autonomous Systems, Special Issue on "Practice and Future of Autonomous Agents." 20 (1997) Nos. 2-4.

.... 3 NL Gr J;v B O D9G K o C F7Z;k 5 l F 0lJ 3 3 tG 9TF07 AI d New AI H8F P l k 7 9 F H4D6 H NAj8 :nMQ KCeL 7 7 7 W m A K 4X 4 , 8 C F k Pfeifer i O 7 9 F H4D6 H NAj8 :nMQ r9MN8 9 k 3 H K h j B 3 K 1 kCN3PLdBj ,Hs o K4JC12= 5 l k 3 H r ( 7 [2][3] H i O HBN d 7 9 F H4D6 H NAj8 :nMQ O I A i b CNG= NH 8= K=EMW JF r 9 k 3 H r gD 7 embodied cognitive science H8F P l k 7 7 J K O r9=C[ 7 h H 7 F k [4] 7 7 0l HL K 7 9 F H4D6 H NAj8 :nMQ rE, Z K 7W 9 k N OHs o K: Fq J a L 3N D8z2LE JCNG= N9=C[K ....

Pfeifer R. and Scheier C.: "Sensory-motor coordination: the metaphor and beyond", Robotics and Autonomous Systems, 20, Special Issue on "Practice and Future of Autonomous Agents", R.Pfeifer and R.Brooks eds, pp.157-178 (1997)

....experiences. Moreover, they disover such categories by acting upon the objects and exploiting the resulting multimodal experiences of objects (i.e. exploration) In our previous work we have developed an approach to adaptive categorization in mobile robots that is based on these results ( 7] [8], 10] 11] The main idea is to view categorization as a sensory motor coordination rather than an isolated perceptual (sub )system. This is achieved by including the robot s own actions into the classification process. In these experiments we have demonstrated how these ideas can be used to ....

....about important properties (e.g. texture, conductivity) of the explored object. A second improvement concerns the categorization mechanisms used. Previously, categorization was based on (a) learning a sensory motor mapping using a temporal Kohonen map [7] or growing dynamical cell structures ([8]) and (b) associating this mapping with behavioral processes such as grasping, pushing or avoiding. The basic categorization mechanism was a conditioned association between the learned sensory motor mappings and some behaviors. In the experiments presented in this paper categorization is not ....

R. Pfeifer and C. Scheier. Sensory-motor coordination: the metaphor and beyond. Journal of Robotics and Autonomous Systems, in press.

.... Similarly, in computer vision systems categorization is seen as a problem of matching the visual input to a stored representation or model of objects (see e.g. 5] for an overview) In our previous work we have developed an alternative approach to categorization adopting the New AI framework ( 8] [11], 12] The main idea is to view categorization as a sensory motor coordination rather than an isolated perceptual (sub )system. This is achieved by including the robot s own actions into the classification process. In this paper we considerably extend this framework. First, we introduce a new ....

R. Pfeifer and C. Scheier. Sensory-motor coordination: the metaphor and beyond. Journal of Robotics and Autonomous Systems, in press.

.... The robot has 8 IR sensors which also function as ambient light sensors, and a gripper with two degrees of freedom (the gripper is not shown in the schema on the left) In previous papers we have shown that the robot can learn these distinctions through so called sensory motor coordination ( 4] [5]) see description of the design principles, below) Whenever the robot manages to pick up a peg, it gets a reinforcement signal. In order to speed up learning, the robot is equipped with reflexes that make it circle around an object if it encounters one. This circling behavior is similar to the ....

....data. This is shown in figure 4. steps (150msec) correlation 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 start SMC end SMC start SMC end SMC Fig. 4: The correlation of the 10 dimensional state space vectors over time, as the agent is moving about in the open and as it is encountering objects (from [5]) The correlation is at an intermediate level as the agent moves about in the open (due to noise) As the agent approaches an object, the correlation drops because now there is rapid change in sensory activation. Once the agent is near the object, the dynamics of the reflexes begins to play and ....

[Article contains additional citation context not shown here]

Pfeifer, R., and Scheier, C. (1997). Sensory-motor coordination: the metaphor and beyond. In R. Pfeifer, and R. Brooks (eds.). Robotics and Autonomous Systems, Special Issue on "Practice and Future of Autonomous Agents." 20, Nos. 2-4.

No context found.

R Pfeifer and C. Scheier, C., "Sensory-motor coordination: the metaphor and beyond" Robotics and Autonomous Systems, (in press).

No context found.

Pfeifer, R. and C. Scheier, "Sensory-motor coordination: the metaphor and beyond," Robotics and Autonomous Systems, Special Issue on "Practice and Future of Autonomous Agents," vol. 20, No. 2-4, pp. 157-178, 1997

No context found.

Pfeifer, R., Scheier, C.: Sensory-motor coordination: The metaphor and beyond. Robotics and Autonomous Systems 20 (1997) 157--178

No context found.

Pfeifer, R. & Scheier, C. (1997). Sensory-motor coordination: the metaphor and beyond. Robotics and Autonomous Systems (special issue on Practice and Future on Autonomous Agents), 20:157-178.

No context found.

Pfeifer, R. and Scheier, C. (1997). Sensory-motor coordination: the metaphor and beyond. Robotics and autonomous systems, Vol. 20, No. 24, pp. 157-178.

No context found.

R. Pfeifer and C. Scheier, "Sensory-Motor Coordination: The Metaphor and Beyond," Robotics and Autonomous Systems, Vol. 20, Nos. 2--4, June 1997, pp. 157--178.

No context found.

R. Pfeifer and C. Scheier, "Sensory-motor coordination: The metaphor and beyond," Robotics and Autonomous Systems, vol. 20, pp. 157-178, 1997.

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