No full-text available
To read the full-text of this research,
you can request a copy directly from the authors.
Interaction and Intelligent Behavior
Abstract
This thesis addresses situated, embodied agents interacting in complex domains. It focuses on two problems: 1) synthesis and analysis of intelligent group behavior, and 2) learning in complex group environments. Basic behaviors, control laws that cluster constraints to achieve particular goals and have the appropriate compositional properties, are proposed as effective primitives for control and learning. The thesis describes the process of selecting such basic behaviors, formally specifying them, algorithmically implementing them, and empirically evaluating them. All of the proposed ideas are validated with a group of up to 20 mobile robots using a basic behavior set consisting of: safe--wandering, following, aggregation, dispersion, and homing. The set of basic behaviors acts as a substrate for achieving more complex high--level goals and tasks. Two behavior combination operators are introduced, and verified by combining subsets of the above basic behavior set to implement collectiv...
... . Swarming and coordinated collective movement Nature presents ample examples of coordinated, collective motion of large swarms of individual organisms: bird flocking, fish schooling, herd stampeding, insect swarming, human pedestrian traffic, and crowd evacuation (Wolff, 1973;Patterson et al., 2007;Moussaïd et al., 2009;Sumpter, 2010;Barnett et al., 2016;Ward and Webster, 2016; see Figure 1). These important natural phenomena have inspired considerable research efforts, ranging from analysis in mathematics, computer science, and physics (Henderson, 1971;Vicsek et al., 1995;Edelstein-Keshet, 2001;Helbing et al., 2001;Giardina, 2008;Vicsek and Zafeiris, 2012), through the development of synthetic swarms in graphics (Reynolds, 1987;Tu and Terzopoulos, 1994) and simulations (Blue and Adler, 2000;Helbing et al., 2001;Daamen and Hoogendoorn, 2003;Toyama et al., 2006;Tissera et al., 2007;Fridman and Kaminka, 2010;Tsai et al., 2011;Kaminka and Fridman, 2018), to robotics (Matari, 1994;Svennebring and Koenig, 2004;Correll and Martinoli, 2009;Kernbach et al., 2010;Mayet et al., 2010;Rubenstein et al., 2012;Brambilla et al., 2013;Giuggioli et al., 2016;LeventBayindir, 2016;Haghighat and Martinoli, 2017;Gauci et al., 2018;Hamann, 2018;Schranz et al., 2020;Dorigo et al., 2021; see Figure 1). ...
... A wide range of theoretical explanations exists for the emergence of collective motion in animals (see reviews in Edelstein-Keshet, 2001;Giardina, 2008;Eftimie, 2012;Vicsek and Zafeiris, 2012;Escobedo et al., 2020), and many attempts have been made to relate these to simulations and robotic swarms (Reynolds, 1987;Parker, 1993;Matari, 1994;Brutschy et al., 2014;LeventBayindir, 2016;Hamann, 2018;Schranz et al., 2020;Dorigo et al., 2021). ...
... There is, however, a separate and distinguished body of research into multi-robot systems that focuses on robotic swarms (Matari, 1994;Kernbach et al., 2010;Mayet et al., 2010;Rubenstein et al., 2012;Brambilla et al., 2013;LeventBayindir, 2016;Haghighat and Martinoli, 2017;Gauci et al., 2018;Hamann, 2018;Schranz et al., 2020;Dorigo et al., 2021), very much in parallel to the above described biological work. It focuses on highly-localized perception and action, leading to numerous local interactions and an emergent order. ...
Swarming or collective motion is ubiquitous in natural systems, and instrumental in many technological applications. Accordingly, research interest in this phenomenon is crossing discipline boundaries. A common major question is that of the intricate interactions between the individual, the group, and the environment. There are, however, major gaps in our understanding of swarming systems, very often due to the theoretical difficulty of relating embodied properties to the physical agents—individual animals or robots. Recently, there has been much progress in exploiting the complementary nature of the two disciplines: biology and robotics. This, unfortunately, is still uncommon in swarm research. Specifically, there are very few examples of joint research programs that investigate multiple biological and synthetic agents concomitantly. Here we present a novel research tool, enabling a unique, tightly integrated, bio-inspired, and robot-assisted study of major questions in swarm collective motion. Utilizing a quintessential model of collective behavior—locust nymphs and our recently developed Nymbots (locust-inspired robots)—we focus on fundamental questions and gaps in the scientific understanding of swarms, providing novel interdisciplinary insights and sharing ideas disciplines. The Nymbot-Locust bio-hybrid swarm enables the investigation of biology hypotheses that would be otherwise difficult, or even impossible to test, and to discover technological insights that might otherwise remain hidden from view.
... Características como o paralelismo, a distribuição ou a descentralização normalmente associadas aos sistemas multi-agente acarretam questões acerca do grau e tipo de interacção estabelecidas entre os intervenientes. Em particular, podem surgir dúvidas acerca de (Mataric 1994): ...
... Em (Mataric 1994) propõe-se como definição para interacção a "influência mútua no comportamento". Consequentemente, uma vez que o estado de cada agente depende dos estados e respectivas acções de cada um dos restantes, o número de estados possíveis cresce exponencialmente num sistema multi-agente. ...
... Os sistemas emergentes têm constituído um relevante domínio de investigação, uma vez que aparentam criar "comportamentos a partir do nada" (Mataric 1994), sendo também designados por auto-organizativos. ...
... Observation-based coordination (OBC) is a key challenge to the multi-agent and multi-robot systems. Increasingly, robots and synthetic agents are being deployed in multi-agent virtual environments for training [1] and entertainment [2] , robotic soc- cer [3], hazardous cleanup tasks [4], formation-maintenance tasks [5,6], and more. Many of these applications rely on agents to coordinate with one another based on their observations of each other [7]. ...
... The inference and decision process are often considered computationally too intense for resource-limited robots. Indeed, there have been many investigations of ways to generate robust and predictable globally-coordinated behavior using agent-local control rules that shortcut the inference and decision processes, (e.g. [6,5]). However, these approaches often require painstakingly hand-crafted control rules and have been applied mostly in spatial coordination tasks (see Section 2 for an in-depth exploration of OBC in the literature). ...
... The key ideas in these is to shortcut the inference and decision making process of SBC by introducing reactive coordination (RC) behaviors that tie specific observations of other agents with actions by the coordinating agent. For instance, Mataric demonstrated that many spatial group behaviors can be achieved by combinations of relatively simple agent-local rules, that directly tie spatial observations (e.g., distance and angle to other robot) with actions that should be taken [6]. Balch investigated methods for reliable execution of group tasks, such as foraging and formation maintenance, using hierarchical reactive behaviors that emphasized coordination by reliance on simple relative observations of other agents [5]. ...
There is a very rich variety of systems of autonomous agents, be it software or robotic agents. In particular, multi-agent systems can include agents that may be part of a team and need to coordinate their actions during their distributed task execution. This coordination requires an agent to observe, i.e., to monitor, the other agents in order to detect a possible coordination failure of the team. Several researchers have addressed the problem of monitoring for single or multiple agent systems and have contributed successful, but mainly application-specific, approaches. In this paper, we aim at contributing a unifying, domain-independent statement of the distributed multi-agent monitoring problem. We define the problem in terms of a pre-defined desirable joint state and an observation-state mapping. Given a concrete joint observation during execution, we show how an agent can detect a possible coordination failure by processing the observation-state mapping and the desirable joint state. To illustrate the generality of our formalism, one of the main contributions of the paper, we represent several previously studied examples within our formalism. We note that basic failure detection algorithms can be computationally expensive. We further contribute an efficient method for failure detection that builds upon an off-line compilation of the principled relations introduced. We show empirical results that demonstrate this effectiveness.
... The leaves of the DAGs are the behavior types' respective outputs: internal state changes for internal behaviors and action primitives for external behaviors. One leaf is illustrated inFigure 2. if (condition) then Behavior(args) if (condition) then Behavior(args) if (condition) then Behavior(args) if (condition) then Behavior(args) if (condition) then Behavior(args) if (condition) then Behavior(args) Behavior(args) Behavior(args) if (condition) then Primitive(args) if (condition) then Primitive(args) if (condition) then Primitive(args) This notion of behavior is consistent with that laid out in [22]. In particular, behaviors can be nested at different levels: selection among lower-level behaviors can be considered a higher-level behavior, with the overall agent behavior considered a single " do-the-task " behavior. ...
... Within the framework presented in [24], the architecture is for interactive software and hardware multi-agents. As mentioned in Section 3, the concept of behavior in the context of our team member agent architecture is consistent with that laid out by Mataric [22]. There, " behavior " is defined as " a control law with a particular goal, such as wall-following or collision avoidance. ...
Multi-agent domains consisting of teams of agents that need to collaborate in an adversarial environment offer challenging research opportunities. In this article, we introduce periodic team synchronization (PTS) domains as time-critical environments in which agents act autonomouslywith low communication, but in which they can periodically synchronize in a full-communication setting. The two main contributions of this article are a flexible team agent structure and a method for inter-agent communication in domains with unreliable, single-channel, low-bandwidth communication. First, the novel team agent structure allows agents to capture and reason about team agreements. We achieve collaboration between agents through the introduction of formations. A formation decomposes the task space defining a set of roles. Homogeneous agents can flexibly switch roles within formations, and agents can change formations dynamically, according to pre-defined triggers to be evaluated at run-ti...
... Reactive layer relies on coupling sensors information and actuators response into a set of low level primitive modules or behaviors. Every behavior deals only with local, short-term information related to the goals or scope of the module, so it doesn't need a complex model of the environment[10]. Reactive Behaviors are very well suited to fast low-level decisions, so this layer is specially important in dynamic unstructured environments. ...
In reactive layers of robotic architectures, behaviors should learn their operation from experience, following the trends of modern intelligence theories. A Case Based Reasoning (CBR) reactive layer could achieve this goal but, as complexity of behaviors increases, the curse of dimensionality arises:too many cases in the behaviors casebases degrade response times so robot’s reactiveness is finally too slow for a good performance. In this work we analyze this problem and propose some improvements in the traditional CBR structure and retrieval phase, at reactive level, to reduce the impact of scalability problems when facing complex behaviors design.
... Usually, this system uses a few fixed base stations to triangulate the position of a robot. However, the bearing of the robot is difficult to obtain in such systems [8]. Some researchers have tried to use simultaneous localization and mapping techniques [9] typically used in single robot setting, but this requires building a map of the area and is typically suitable where obstacles can provide identifiable patterns in the environment. ...
Self-localization, a process by which a robot obtains its own global positional state (consisting of position and bearing), is an important aspect of multi-robot experiments. A traditional method to carry Out self-localization is to attach a unique visual pattern to the robot. A centralized sensor, such as a camera, is typically used to identify the pattern and measure its state directly. However, the use of patterns presents several challenges and limitations. To eliminate the use of a pattern, a novel method has been proposed in this paper for the robots to obtain their global states by themselves. In this method, we implement an extended Kalman filter (EKF) on each robot to fuse the global position data with the control input data to estimate the bearing. In a multi-robot setting, this becomes challenging because the positional data obtained from sensors such as a camera is untagged and robots do not have a prior knowledge about which data pertains to their own position. To overcome this issue, we propose a method in which each robot can identify its own track from the other's tracks by comparing the average measurement residual of the EKFs that are run on each candidate track. Therefore, instead of identifying and localizing robots in a centralized manner, we distribute the task to each robot and let them localize themselves. Extensive experiments have been conducted and the results are provided to show the effectiveness of this method.
... [4] [18] [20]. Another solution consists of using topological maps, in which the environment is represented by means of objects and graphs; [6] [7] [9] [14] [16] path planning becomes easier, but learning and sensor integration becomes more difficult. ...
... Within the field of robotics, there has been considerable research into multirobot coordination for a variety of domains and tasks (e.g., [3, 11, 13, 21, 25, 34] ). In particular, there has been an ongoing effort to explore multirobot teamwork in dynamic, adversarial tasks within a standardized domain: RoboCup robot soccer [2, 18, 27, 33] (see also http://www.robocup.org), ...
In this paper, we focus on human-robot interaction in a team task where we identify the need for peer-to-peer (P2P) teamwork, with no fixed hierarchy for decision making be- tween robots and humans. Instead, all team members are equal participants and decision making is truly distributed. We have fully developed a P2P team within Segway Soccer, a research domain, built upon Robocup robot soccer, that we have introduced to explore the challenge of P2P coor- dination in human-robot teams with dynamic, adversarial tasks. We recently participated in the first Segway Soccer games between two competing teams at the 2005 RoboCup US Open. We believe these games are the first ever between two human-robot P2P teams. Based on the competition, we realized two different approaches to P2P teams. We present our robot-centric approach to P2P team coordination and contrast it to the human-centric approach of the opponent team.
... Consequently, over the last decade, there has been a urry of work on map building for mobile robots (see e.g., 7, 36, 54, 67]). Some approaches seek to devise abstract, topological descriptions of robot environments 1, 8, 33, 42], whereas others generate more detailed, metric maps 7, 15, 22, 39, 44]. Naturally, the problem of mapping lends itself nicely to multi-robot solutions, where multiple robots collaborate and jointly explore an unknown environment. ...
An efficient probabilistic algorithm for the concurrent mapping and localization problem that arises in mobile robotics is presented. The algorithm addresses the problem in which a team of robots builds a map on-line while simultaneously accommodating errors in the robots’ odometry. At the core of the algorithm is a technique that combines fast maximum likelihood map growing with a Monte Carlo localizer that uses particle representations. The combination of both yields an on-line algorithm that can cope with large odometric errors typically found when mapping environments with cycles. The algorithm can be implemented in a distributed manner on multiple robot platforms, enabling a team of robots to cooperatively generate a single map of their environment. Finally, an extension is described for acquiring three-dimensional maps, which capture the structure and visual appearance of indoor environments in three dimensions.
... Hoff [3] designed algorithms to learn both the basic behaviors with sub-goals and how to combine them to achieve more complex behaviors. Mataric [7] gave an impressive demonstration of behavior-based control on real robots. Two behaviors can be combined into new higher-level behaviors by behavior combination operators. ...
Teams of robots can be redundant with respect to a given task. This redundancy can be exploited to pursue additional objectives during the execution of the task. In this paper, we describe a control-based method to exploit such redundancy for the execution of additional behavior, leading to the improvement of overall performance. The control-based method provides a suitable mechanism for combining controllers with different objectives. The mechanism ensures that the subordinate controllers do not interfere with the superior controllers. Thus it allows to build controllers exhibiting complex behavior from simple primitives, while maintaining their provable performance characteristics. The effectiveness of the framework is demonstrated by experiments with a multi-robot exploration task.
... The leaves of the DAGs are the behavior types' respective outputs: internal state changes for internal behaviors and action primitives for external behaviors. Our notion of behavior is consistent with that laid out in 4]. In particular, behaviors can be nested at diierent levels: selection among lower-level behaviors can be considered a higher-level behavior, with the overall agent behavior considered a single \do-the-task" behavior. ...
The Soccer Server system provides a rich and cha]]enging multiagent, real-time domain. Agents must accurately perceive and act despite a quickly changing, largely hidden, noisy world. They must also act at several ]eve]s, ranging from individual skills to full-team collaborative and adversarial behaviors. This article presents the CMUnited-97 approaches to the above challenges which helped the team to the semifinals of the 29-team RoboCup-97 tournament.
... Other definitions noted are " joint collaborative behaviour that is directed toward some goal in which there is a common interest or reward " [Barnes and Gray, 1991]; " a form of interaction, usually based on communication " [Matariþ, 1994a]; and " [joining] together for doing something that creates a progressive result such as increasing performance or saving time " [Yuta and Premvuti, 1990]. These three definitions reflect emphasis on task, mechanism and performance respectively. ...
This thesis develops the Architecture for Behaviour Based Agents (ABBA) -- an architecture suitable for supporting the distributed planning of cooperative behaviour in multi-robot systems. ABBA was used to implement a concrete cooperative cleaning task using two mobile robots, both to drive the design requirements and as a demonstration of its efficacy. The cleaning task solution requires reactive and deliberative behaviour, purposive navigation by learning unknown environments, cooperation and communication. Learning to navigate purposively in an unknown environment obviously requires a map. In adherence with the behaviour-based philosophy, the navigation mechanism developed uses no explicit representation. The map representation arises as a natural consequence of the action selection dynamics, correlation learning and the key notion of feature detectors for `locations'. The robot learns the spatial and topological adjacency of locations through behaving. The implementation of this m...
... [4] [18] [20]. Another solution consists of using topological maps, in which the environment is represented by means of objects and graphs; [6] [7] [9] [14] [16] path planning becomes easier, but learning and sensor integration becomes more difficult. ...
In this paper a system for estimating position and orientation of a mobile robot in a non-structured environment, by means of self-organizing neural networks, is described. After performing an initial learning phase, the system is able to incrementally build a representation of the environment. Position and orientation of the robot can then be obtained, upon data coming from the odometric system and sensors (ultrasound sensors, infrared sensors and 2D-laser sensor). The system has been implemented and tested on a NOMAD 200 robot. Experimental results are reported, which show the robot to be able to navigate without losing its way, and to adapt to changes in the environment. 1.-
... Intelligent autonomous robots perform tasks according to different behaviors [4, 9, 1, 12]. We can assume that in general, each robot will act according to a set of behaviors , either fully or partially pre-defined. ...
Automated recognition of the behavior of robots is increasingly needed in a variety of tasks, as we develop more autonomous robots and general information processing agents. For example, in environments with multiple autonomous robots, a robot may need to make decisions based on the behavior of the other robots. As another interesting example, an intelligent narrator agent observing a robot will need to automatically identify the robot's behaviors. In this paper, we introduce a novel framework for using Hidden Markov Models (HMMs) to represent and recognize strategic behaviors of robotic agents. We first introduce and characterize the perceived signal in terms of behavioral-relevant state features. We then show how several HMMs capture different defined robot behaviors. Finally we present the HMM-based recognition algorithm which orchestrates and selects the appropriate HMMs in real time. We use the multi-robot robotic soccer domain as the substrate of our empirical validation, both in...
Within the field of multi-agent systems (MASs), communication is often viewed as an abstract and dyadic process. In this thesis, we argue that MASs can profit from overhearing, a form of communication found in human activity that can be used to give robustness and flexibility to their structure.
To this end, we propose the star model, which allows a MAS to use overhearing to dynamically adopt a structure based on leader-centered groups, following the evolution of limited and dynamic communication possibilities. This model provides a framework for the agents' activities that imposes constraints on the agents' decision processes (most notably coherence), but allows them to use every message that they receive (even if they are not the intended recipients) to update their representations of the MAS organization, and thereby facilitating the overall activity of the MAS.
On the one hand, we develop several elements in order to exploit this model (a language with associated semantics, an update mechanism and an algorithm). On the other hand, two series of experiments (OTTO and LOTTO) allow us to study the effects of the variation of several parameters on the behavior of the model.
Globalization of markets, embodiment of new technologies, and the need of fast reconfigurability are leading small manufacturing enterprises into changing their manufacturing systems for adaptable and flexible ones. In this context, IT agentification is the key of a new approach for adapting obsolete manufacturing systems to the requirements of the new economy. In this paper we present a theoretical model, and our practical experiences in the agentification of a small manufacturing plant.
We describe an empirical study that sought general guidelines for
task allocation strategies in multi-robot systems. We identify four
distinct task allocation strategies, and demonstrate them in two
versions of the multi-robot emergency handling task. We describe an
experimental setup to compare results obtained from a simulated grid
world to the results from real world experiments. Data resulting from
eight hours of real mobile robot experiments are compared to the trend
identified in simulation. The data from the simulations show that there
is no single strategy that produces best performance in all cases, and
that the best task allocation strategy changes as a function of the
noise in the system. This result is significant, and shows the need for
further investigation of task allocation strategies
This demonstration showcases distributed algorithms for configuration control in multi-robot systems. These algorithms include examples of gradient communication, clustering and dispersion, group motion, and network characterization. The algorithms are demonstrated on a swarm of 15 mobile robots.
In the quest for machines that are able to learn, reinforcement
learning (RL) is found to be an appealing learning methodology. A known
problem in this learning method, however is that it takes too long
before the robot learns to associate suitable situation-action pairs.
Due to this problem, RL has remained applicable only to simple tasks and
discrete environment. To accelerate the learning process to a level
required by real robot tasks, the traditional learning architecture has
to be modified. We propose a modified reinforcement based robot skill
acquisition and adaptation architecture. The architecture has two
components: a bias and a learning components. The bias component imparts
to the learner coarse a priori knowledge about the task. Subsequently,
the learner refines the acquired actions through reinforcement learning.
We have validated the architecture and the learning algorithm on a
simulated TRC mobile robot for a goal reaching task
Reinforcement learning agents attempt to learn and construct a decision policy which maximises some reward signal. In turn, this policy is directly derived from long-term value estimates of state-action pairs. In environments with real-valued state-spaces, however, it is impossible to enumerate the value of every state-action pair, necessitating the use of a function approximator in order to infer state-action values from similar states. Typically, function approximators require many parameters for which suitable values may be difficult to determine a-priori. Traditional systems of this kind are also then bound to the fixed limits imposed by the initial parameters, beyond which no further improvements are possible. This paper introduces a new method to adaptively increase the resolution of a discretised action-value function based upon which regions of the state-space are most important for the purposes of choosing an action. The method is motivated by similar work by Moor...
This paper proposes an approach for the development of multi-agent architectures based on a conceptual model set which is implementation independent. We propose an extension of the CommonKADS methodology to multi-agent systems; run-time aspects are taken into consideration, in particular, the structuring of problem-solving knowledge at a conceptual level guarantees flexibility and modularity and an execution level. The construction of agent models is a compositional process based on the CommonKADS-close conceptual models resulting from five analysis steps. Inference structures have appeared to be a useful means to guide the implementation of our competence-encapsulating agent model. In fact, we use them to identify procedures from a program library during implementation.
This work demonstrates the application of the behavior--based approach to generating ethologically--inspired adaptive foraging using a division of labor into exclusive spatial territories. First, we use fixed group sizes to evaluate and compare the performance of the two types of adaptive solutions. Second, using a collection of experimental robot data, we empirically derive and demonstrate the critical mass for most effective foraging in our domain, and show the decline of performance of the space division strategy with increased group size. 1 Introduction This paper describes work that applies the behavior-- based approach for autonomous agent control to generating ethologically--inspired adaptive foraging. In contrast to foraging methods explored by various researchers to date (see section 3 for an overview), we employ a territorial principle that implements a division of labor into exclusive spatial areas. First, we use fixed group sizes of two, three, and four robots ...
Distributed Artificial Intelligence (DAI) has existed as a subfield of AI for less than two decades. DAI is concerned with systems that consist of multiple independent entities that interact in a domain. Traditionally, DAI has been divided into two sub-disciplines: Distributed Problem Solving (DPS) focuses on the information management aspects of systems with several components working together towards a common goal; Multiagent Systems (MAS) deals with behavior management in collections of several independent entities, or agents. This survey of MAS is intended to serve as an introduction to the field and as an organizational framework. A series of general multiagent scenarios are presented. For each scenario, the issues that arise are described along with a sampling of the techniques that exist to deal with them. The presented techniques are not exhaustive, but they highlight how multiagent systems can be and have been used to build complex systems. When options exist, the tec...
A theoretical analysis of the cognitive capacities underlying self-awareness and other-awareness suggested that (a) self- and other-awareness are closely linked because both require a cognitive capacity for secondary representation, and (b) other-awareness facilitates the synchronic imitation of object use with an unfamiliar peer. The relation between mirror self-recognition and synchronic imitation was investigated in 56 dyads of unfamiliar 19-month-old children who were systematically paired according to their mirror self-recognition status and who were observed during free play with sets of duplicate toys. Long phases of synchronic imitation occurred in nearly all dyads consisting of recognizers but in only one dyad composed of nonrecognizers; in mixed dyads these phases were shorter than in recognizer dyads. Discussion focuses on the synchrony of the development of self- and other-awareness.
How are we able to understand and anticipate each other in everyday life, in our daily interactions? Through the use of such "folk" concepts as belief, desire, intention, and expectation, asserts Daniel Dennett in this first full-scale presentation of a theory of intentionality that he has been developing for almost twenty years. We adopt a stance, he argues, a predictive strategy of interpretation that presupposes the rationality of the people - or other entities - we are hoping to understand and predict.These principles of radical interpretation have far-reaching implications for the metaphysical and scientific status of the processes referred to by the everday terms of folk psychology and their corresponding terms in cognitive science.While Dennett's philosophical stance has been steadfast over the years, his views have undergone successive enrichments, refinements, and extensions. "The Intentional Stance" brings together both previously published and original material: four of the book's ten chapters - its first and the final three - appear here for the first time and push the theory into surprising new territory. The remaining six were published earlier in the 1980s but were not easily accessible; each is followed by a reflection - an essay reconsidering and extending the claims of the earlier work. These reflections and the new chapters represent the vanguard of Dennett's thought. They reveal fresh lines of inquiry into fundamental issues in psychology, artificial intelligence, and evolutionary theory as well as traditional issues in the philosophy of mind.Daniel C. Dennett is Distinguished Arts and Sciences Professor at Tufts University and the author of "Brainstorms" and "Elbow Room." "The Intentional Stance," along with these works, is a Bradford Book.
This article describes an architecture for cooperative and au tonomous mobile robots. The architecture is composed of three levels: functional, control, and planner levels. The functional and control levels manage robot resources, while the planner level carries out complex operations such as planning and cooperation.
The cooperation is composed of two phases: (1) collabora tion, in which a task is decomposed into subtasks and subtasks are allocated through a set of robots; and (2) coordination, in which robots actually coordinate their activities to fulfill the initial task using the notion of coordinated protocols. Moreover, a language has been developed to program the coordination.
This architecture exhibits important benefits such as (1) mod ularity (the robot can work autonomously or within a team), (2) robustness (although some modules on the robot fail, it is still able to perform useful tasks), and (3) programmability. Two actual examples using two indoor mobile robots are described to demonstrate the interest of this approach.
Studied deferred imitation of object-related actions in 82 12–13 mo olds and in 85 18–29 mo olds to examine the development of competence in observational learning. Three task categories were investigated: simple/single actions, reiterative actions, and sequentially coordinated actions. Results show evidence for deferred imitation for the simple/single actions at 12 mo. When the simple/single action component was isolated in the more complex tasks and its achievement scored, deferred imitation was determinable for additional actions. At 18 mo, improvements occurred in observational learning for all 3 task categories. However, even at 18 mo, serial and sequential coordinations were accomplished by fewer than half of the Ss who had observed actions with these properties. Examination of partial successes at both ages suggested phases in the information sequence. In these phases, encoding was probably inadequate, and action schemes were probably insufficiently differentiated for Ss to fully profit from modeling. Findings suggest that imitative ability is in an early developmental phase by the end of the 1st yr and that information-processing characteristics have more demonstrable effects during the 2nd yr of life. (24 ref) (PsycINFO Database Record (c) 2012 APA, all rights reserved)
Discrete Event Systems (DES) are a special type of dynamic systems. The "state" of these systems changes only at discrete instants of time and the term "event" is used to represent the occurrence of discontinuous changes (at possibly unknown intervals). Different Discrete Event Systems models are currently used for specification, verification, synthesis as well as for analysis and evaluation of different qualitative and quantitative properties of existing physical systems.
The main focus of this paper is the presentation of the automata and formal language model for DES introduced by Raniadge and Wonham in 1985. This model is suitable for the examination of some important control theoretic issues, such as controllability and observability from the qualitative point of view, and provides a good basis for modular synthesis of controllers. We will also discuss an Extended State Machine and Real-Time Temporal Logic model introduced by Ostroff and Wonham in [OW87]. It incorporates an explicit notion of time and means for specification and verification of discrete event systems using a temporal logic approach. An attempt is made to compare this model of DES with other ones.
There are two main published models for flower-shape memory, for landmark memory, and for locale memory in honey bees. Recent experiments have shed light on the actual mechanisms in each of these cases. Both shape and landmark memory are pictorial, as opposed to parameter-based; they have different resolutions, and so are probably stored separately; neither is eidetic. Locale memory involves a cognitive map rather than just route-specific memory. One of the most interesting questions about learning is how information is processed and stored. This paper will review recent progress in the understanding of how honeybees store and use visual information gathered on foraging flights.
An architecture that integrates a map representation into a
reactive, subsumption-based mobile robot is described. This fully
integrated reactive system removes the distinction between the control
program and the map. The method was implemented and tested on a mobile
robot equipped with a ring of sonars and a compass, and programmed with
a collection of simple, incrementally designed behaviors. The robot
performs collision-free navigation, dynamic landmark detection, map
construction and maintenance, and path planning. Given any known
landmark as a goal, the robot plans and executes the shortest known path
to it. If the goal is not reachable, the robot detects failure, updates
the map, and finds an alternate route. The topological representation
primitives are designed to suit the robot's sensors and its navigation
behavior, thus minimizing the amount of stored information. Distributed
over a collection of behaviors, the map itself performs constant-time
localization and linear-time path planning. The approach is qualitative
and robust
In this paper we describe the subsumption architecture, a design methodology for multiprocessor control systems. The architecture is based on a new decomposition of problems into task-achieving behaviors. This decomposition allows us to incrementally construct more competent robots by adding new behaviors. We demonstrate this by showing how two different control systems can be built on top of the same set of core behaviors. We have implemented both systems and have run them on a real mobile robot. The details and performance of each will be discussed.
We describe an algorithm which allows a behavior-based robot to learn on the basis of positive and negative feedback when to activate its behaviors. In accordance with the philosophy of behavior-based robots, the algorithm is completely distributed: each of the behaviors independently tries to sensors find out (i) whether it is relevant (i.e. whether it is at all correlated to positive feedback) and (ii) what the conditions are under which it becomes reliable (i.e. the conditions under which it maximises the probability of receiving positive feedback and minimises the probability of receiving negative feedback). The algorithm has been tested successfully on an autonomous 6-legged robot which had to learn how to coordinate its legs so as to walk forward.
We investigate the effect of predictions upon a model of coevolutionary systems which was originally inspired by computational ecosystems. The model incorporates many of the features of distributed resource allocation in systems comprised of many individual agents, including asynchrony, resource contention, and decision-making based upon incomplete knowledge and delayed information. Previous analyses of a similar model of non-predictive agents have demonstrated that periodic or chaotic oscillations in resource allocation can occur under certain conditions, and that these oscillations can affect the performance of the system adversely. In this work, we show that the system performance can be improved if the agents do an adequate job of predicting the current state of the system. We explore two plausible methods for prediction - technical analysis and system analysis. Technical analysts are responsive to the behavior of the system, but suffer from an inability to take their own behavior into account. System analysts perform extremely well when they have very accurate information about the other agents in the system, but can perform very poorly when their information is even slightly inaccurate. By combining the strengths of both methods, we obtain a successful hybrid of the two prediction methods which adapts its model of other agents in response to the observed behavior of the system.
How can the semantic interpretation of a formal symbol system be made
intrinsic to the system, rather than just parasitic on the meanings in our
heads? How can the meanings of the meaningless symbol tokens, manipulated
solely on the basis of their (arbitrary) shapes, be grounded in anything but
other meaningless symbols? The problem is analogous to trying to learn Chinese
from a Chinese/Chinese dictionary alone. A candidate solution is sketched:
Symbolic representations must be grounded bottom-up in nonsymbolic
representations of two kinds: (1) "iconic representations," which are analogs
of the proximal sensory projections of distal objects and events, and (2)
"categorical representations," which are learned and innate feature-detectors
that pick out the invariant features of object and event categories from their
sensory projections. Elementary symbols are the names of these object and event
categories, assigned on the basis of their (nonsymbolic) categorical
representations. Higher-order (3) "symbolic representations," grounded in these
elementary symbols, consist of symbol strings describing category membership
relations (e.g., "An X is a Y that is Z").
A method for investigating the behavioral sequences used in forming dominance hierarchies is presented. There are 4 possible sequences for the formation of the first 2 dominance relationships in groups of 3 individuals (triads). Designating the winner in the first relationship formed as the initial dominant, the loser as the initial subordinate and the animal not involved as the bystander, the 4 possible sequences are: 1) the initial dominants goes on to dominate the bystander (Double Dominance), 2) the bystander later dominates the initial subordinate (Double Subordinance), 3) the bystander later dominates the initial dominant (Bystander Dominates Initial Dominant), and 4) the initial subordinate later dominates the bystander (Initial Subordinate Dominates Bystander). Although each sequence has an equal probability of occurrence if dominance relationships are formed randomly, 2 of the sequences have different implications for the formation of the empirically common linear and near linear hierarchies than the other two. Sequences of dominance relationships formed in groups of 3 and 4 chickens were analyzed. In both experiments Double Dominance and Double Subordinance composed the overwhelming majority of all sequences - 91% in triads and 87% in tetrads. -from Author
An effective method to create an autonomous reactive controller is to learn a model of the environment and then use dynamic programming to derive a policy to maximize long term reward. Neither learning environmental models nor dynamic programming require parametric assumptions about the world, and so learning can proceed with no danger of becoming “stuck― by a mismatch between the parametric assumptions and reality. The paper discusses how such an approach can be realized in real valued multivariate state spaces in which straightforward discretization falls prey to the curse of dimensionality.
Research on mobile robots began in the late sixties with the Stanford Research Institute’s pioneering work. Two versions of SHAKEY, an autonomous mobile robot, were built in 1968 and 1971. The main purpose of this project was “to study processes for the realtime control of a robot system that interacts with a complex environment” 〈NIL 69〉. Indeed, mobile robots were and still are a very convenient and powerful support for research on artificial intelligence oriented robotics. They possess the capacity to provide a variety of problems at different levels of generality and difficulty in a large domain including perception, decision making, communication, etc., which all have to be considered within the scope of the specific constraints of robotics: on-line computing, cost considerations, operating ability, and reliability.
A phenomenological model for osmotropotactic behaviour in ants and termites is presented. The model concentrates on the physical chemistry of scent trails and the dynamics between a chemical stimulus and the behavioural response at the level of the individual using a non-linear perception-response function. The model reproduces experimental data and shows that a flexible response, saturation at the level of perception and a certain amount of noise are necessary to obtain trail following. The comparison between theoretical and experimental results reveals what factors are important in general for orientation on scent trails. The same model also describes orientation towards point sources. Various examples for collective patterns, which are based on trail following or aggregation, are known in ants and termites. These patterns emerge from the integration of individual behaviour taking into account interactions between the individuals. The model presented can be used for a group of equally equipped individuals and serves thus as a starting point for the description of complex behaviour.
Abstract In recent years, there has been a growing fascination with decentralized systems and self-organizing phenomena. Increasingly, people are choosing decentralized models for the organizations and technologies that they construct in the world, and for the theories that they construct about the world. But even as decentralizedideas spread through the culture, there is a deep-seated resistance to such ideas. In trying to understand patterns in the world, people often assume centralized control where none exists (for example, assuming that a "leader bird" guides the rest of the flock). To probe how people think about decentralized systems, and to help them develop new ways of thinking about such systems, I developed a programmable modeling environment (called StarLogo) with which people can easily create and experiment with decentralized systems. StarLogo allows users to control the actions and interactions of thousands of artificial "creatures" on the computer screen. I describe three StarLogo projects created by high-school students. Based on my observations of these (and other) students, I analyze the nature of the centralized mindset, and I discuss how people, through engagement with new types of computationaltools and activities, can begin to move beyond the centralized mindset.
A higher level language derives its great power from the fact that it tends to impose structure on the problem solving behavior of the user. Besides providing a library of useful subroutines with a uniform calling sequence, the author of a higher level language imposes his theory of problem solving on the user. By choosing what primitive data structures, control structures, and operators he presents, he makes the implementation of some algorithms more difficult than others, thus discouraging some techniques and encouraging others. So, to be good, a higher level language must not only simplify the job of programming, by providing features which package programming structures commonly found in the domain for which the language was designed, it must also do its best to discourage the use of structures which lead to bad algorithms.
The unified society of an Apis mellifera colony may be viewed as a functional unit above the level of the genes, which operates to enhance their survival and reproduction. The completeness of cooperation in the colony may be measured by the degree of congruence in the genetic interests of a colony's members, and colonies containing a mother queen may be regarded as nearly true superorganisms. The author focuses on the proximate mechanism by which colonies function as integrated wholes by examining the architecture of information flow. The key to this is studies of worker activities, viz the large amount of time spent in patrolling, in turn related to the gathering of information through cues (information carried only incidentally), signals (stimuli conveying information on food sources, etc) and the shared environment. An example of foraging shows how natural selection has linked pathways of information flow to build up the colony's coordinated group action. -J.W.Cooper
Our book examines the mechanisms that underlie social behavior and communication in East African vervet monkeys. Our goal is to describe the sophistication of primate intelligence and to probe its limits. We suggest that vervets and other primates make good primatologists. They observe social interactions, recognize the relations that exist among others, and classify relationships into types. Monkeys also use sounds to represent features of their environment and compare different vocalizations according to their meaning. However, while monkeys may use abstract concepts and have motives, beliefs, and desires, their mental states are apparently not accessible: they do not know what they know. In addition, monkeys seem unable to attribute mental states to others: they lack a "theory of mind." Their inability to examine their own mental states or to attribute mental states to others severely constrains their ability to transmit information or to deceive one another. It also limits the extent to which their vocalizations can be called semantic. Finally, the skills that monkeys exhibilt in social behavior are apparently domain specific. For reasons that are presently unclear, vervets exhibit adaptive specializations in social interactions that are not extended to their interactions with other species (although they should be).
present some data on the development of eye-contact patterns in an infant gorilla interacting with human adults in problem-solving situations / the subject of this longitudinal study was a hand-reared female infant gorilla . . . observed in a zoo nursery environment from about six months to 30 months (PsycINFO Database Record (c) 2012 APA, all rights reserved)
Shows how a system consisting of 2 neuronlike adaptive elements can solve a difficult control problem in which it is assumed that the equations of the system are not known and that the only feedback evaluating performance is a failure signal. This evaluative feedback is of much lower quality than is required by standard adaptive control techniques. It is argued that the learning problems faced by adaptive elements that are components of adaptive networks are at least as difficult as this problem. The learning system consists of a single associative search element (ASE) and a single adaptive critic element (ACE). In the course of learning to balance the pole, the ASE constructs associations between input and output by searching under the influence of reinforcement feedback, and the ACE constructs a more informative evaluation function than reinforcement feedback alone can provide. The differences between this approach and other attempts to solve problems using neuronlike elements are discussed, as is the relation of the ACE/ASE system to classical and instrumental conditioning in animal learning studies. Implications for research in the neurosciences are noted. (42 ref) (PsycINFO Database Record (c) 2012 APA, all rights reserved)