P. Maes, `How to do the right thing', Connection Science Journal, 1(3), (1990).  Home/Search   Document Details and Download   Summary   Related Articles   Check

....the eyes. Gestures generated by the MConBeh module must be decoded into motor actions to get the desired face expression. MConBeh: The core of the perceptual e#ector system, it is in charge of the generation of the conductual sequences. This module will be based on Behavior Activation Networks [14], which allow to define relations between behaviors and their activation and activated variables as STRIPS like rules. It also resolves the behavior transition problem and gives a simple and complete solution to the action selection problem. MInt: This module serves as interface with the ....

....velocities. A snapshot of the main application window is shown in Figure 5. Figure 5: Pose Editor. 3. 5 Action selection module A software tool called ZagaZ [11] was developed for designing, testing and running PHISH Nets [18] an improved version of Maes Behavior Activation Networks [14]. It provides an easy and scalable way to create behavior networks that can accept external symbolic inputs and generate simple output signals. Also, a debugger was included that allows the implementer to test the designed network. Network modules (normally simple behaviors) can also be other ....

P. Maes. How to do the right thing. Connection Science Journal, 1(3):291--323, 1989.

....them. Our model is able to simulate such situations, without high computational cost. Finally, an autonomous agent has to select its actions by itself. Research has been driven by people from different areas: ethologists such as Tinbergen [20] and computer scientists such as Brooks [6] Maes [13] and Minsky [14] who lead the school of Behaviour Based Artificial Intelligence (BBAI) Our model, as proposed in the BBAI, does not attempt to build models of the world, and the agent has to reevaluate its course of action on every slot of time. Some points are not directly addressed by the BBAI ....

P. Maes. How to do the right thing. Connection Science Journal, 1:291-323, Dec 1989.

....modeled virtual marine world. Hodgins et al. 8] described dynamic athletic behaviors. Unuma et al. 9] modeled human figure locomotion with emotions. Other papers presented behavioral architectures. Brooks [10] 11] developed a subsumption architecture built in layers of behaviors. Maes [12] presented an action selection model where action selection is an emergent property of activation inhibition dynamics among actions. Beer et al. 13] developed neural networks for insects. Ahmad et al. 14] described hierarchical concurrent state machines for behavior modeling and scenario ....

Maes, P.. "How to Do the Right Thing , Connection Science, Vol. 1, No. 3, 1989

....intend to refer to properties of architectures that are functionally specified, but not in terms of components or the structural layout of the architecture. A typical example of an externally specified function would be the selection algorithm for a winner take all action selection scheme (e.g. [8]) This algorithm determines the component with the highest activation out of a set of components with activation levels and passes control to it. Typically, this algorithm will be specified for a set of components, but will not have an architectural representation (neither explicit in terms of a ....

.... result of some competition process among different components, possibly followed by the arbitration of the current behavior (if an action different from the current one was selected during competition) Examples are priority based [5] state based [7] and winner take all competition mechanisms [8]. Using the two way categorization from the previous section, we can further divide competitive and cooperative action selection into three subcategories, respectively (external mechanisms are, by nature, explicit) Tables 1 and 2 are a sample classification of existing architectures under the ....

[Article contains additional citation context not shown here]

P. Maes. How to do the right thing. Connection Science Journal, 1:291--323, 1989.

....are the most general means through which a node becomes active. Activations are determined by the values passed via incoming activation links. How these activation link values are used to compute the final activation of a node is a decision to be made on a case by case basis (e.g. as in [Maes1989] In most cases the simple and straightforward addition of incoming values will be the best way to determine the activation of a node. The second type of link is a priority link (P link) Unlike activation values, which are used in a combinatory fashion to determine an overall best action, ....

Maes, P. 1989. How to do the right thing. Connection Science Journal 1:291--323.

....to unpredictable environmental changes, but these systems seem short sighted when pursuing goals and experience coherency problems due to perceptual restrictions. These problems are well known, and there have been attempts to build architectures which incorporate the best of both worlds [Fir87, Mae90]. With respect to building believable, life like creatures with which humans can empathize, past work in both fields of AI has failed to identify the single most important reason for incorporating an expectation mechanism: without expectations, animals cannot make mistakes . So why in the world ....

....sensors and actuators can be more e#ective than complex planning mechanisms, while exhibiting many of the same capabilities. Examples of these approaches include the Pengi system of Agre and Chapman [AC87] the subsumption architecture of Brooks [Bro86] the spreading activation networks of Maes [Mae90], and the Society of Mind theories of Minsky [Min86] In an attempt to leverage the advantages of both approaches, some hybrid systems like that of Firby [Fir87] have used a planner to make high level behavioral decisions while using a reactive system for low level control during behavior ....

Pattie Maes. How to do the right thing. Connection Science Journal, 1(3):293-- 325, 1990.

....is the final actuation calculated (action selection) There are two major paradigms: arbitration and command fusion. Arbitration establishes a competition for control among all the behaviors and only the winning one determines the final actuation. Priorities, activation networks from Pattie Maes [13] and state based arbitration [3] fall in this category. Command fusion techniques merge all the relevant outputs in a global one that take into account all behavior preferences. Relevant approaches in command fusion include superposition[2] fuzzy blending [15] and voting [14] Coordination is ....

Maes, P.: How to Do the Right Thing. Connection Science Journal (Special Issue on Hybrid Systems), 1(3) (1989) 291--323

....whether a bundle is active or non active, each Behavior agent is endowed with a state. Each bundle maintains whether it is active or non active. It activates itself when a user starts its corresponding activity, and deactivates itself when the user stops. There are other activation models. [27] presents an action selection algorithm for activating agents. These agents, like Brooks robots, are composed of a hierarchy of competence modules. Like our bundles, these modules activate and inhibit each other through various 37 relationships: successor links, predecessor links, and conflictor ....

....our discrete state model, these modules possess a continuous quantity of activation energy. A module receives energy when one of its predecessors receives energy or is activated; it loses energy when a predecessor loses energy or is deactivated. The continuous model is necessary for the system in [27] because its modules sense and react. In the Intelligent Room, the notion of energy levels is encoded in our perceptual systems. Our activation model needs discrete logic for intelligibility; the system needs to know whether a behavior bundle is active or not, not a level of activation. 4.1.4 ....

Pattie Maes. How to Do the Right Thing. Connection Science Journal, 1(3), 1989.

....and combine sequentially to solve the main problem [Singh, 1992, Tham and Prager, 1994] The Action Selection problem essentially concerns subtasks acting in parallel, and interrupting each other rather than running to completion. Typi cally, each subtask can only ever be partially satisfied [Maes, 1989]. 3.1 Hierarchical Q learning Lin has devised a form of multi module RL suitable for such problems, and this will be the second method tested below. Lin [Lin, 1993] suggests breaking up a complex problem into sub problems, having a collection of Q learning agents A1, An learn the ....

....agent is generally in charge, but will be corrected by the other agents whenever it offends them too much. W learning tries to keep everyone on board, while still going somewhere. For a further analysis of how Minimize the Worst Unhappiness gets tasks completed (enforces persistence) see 15.1.3. [Maes, 1989] lists desirable criteria for action selection schemes, and in it we see this tension between wanting actions that contribute to several goals at once and yet wanting to stick at goals until their conclusion. We can represent this in a diagram of single mindedness (Figure 14.1) 116 fanatical ....

[Article contains additional citation context not shown here]

Maes, Pattie (1989), How To Do the Right Thing, Connection Science 1:291- 323.

....paper, we also attempt to combine several neural networks for solving this problem. Each neural network can be evolved or programmed. Evolved neural network is based on CAM Brain model, and a programmed module controls the robot directly. We apply Pattie Maes s Action Selection Mechanism (MASM) [10] to combine modules and control a mobile robot in simulated environments. The rest of this paper introduces CAM Brain model and basic behaviors, and presents the integration method in detail. The detailed description of simulation follows, and the results of simulation is given. 2. Neural ....

P. Maes, "How to do the right thing," Connection Science Journal, Vol 1, No. 3, pp 291-323, 1989.

No context found.

P. Maes, `How to do the right thing', Connection Science Journal, 1(3), (1990).

No context found.

Maes, P. (1989). How to do the right thing. Connection Science Journal, 1(3):291--323.

No context found.

Maes, P. 1989. How to Do the Right Thing. Connection Science 1:3, pp. 291-323.

No context found.

Maes, P.: How to do the right thing. Connection Science Journal 1(3) (1989) (291-- 323)

No context found.

Maes, P. (1989). How to do the right thing. Connection Science Journal, 1.

No context found.

Maes, P.: How To Do the Right Thing. Connection Science Journal 1 (1989) 291--323

No context found.

P. Maes, \How to do the right thing?" Connection Science, 1(3), 1989, pp. 291-323. 35

No context found.

Pattie Maes, How to do the right thing, AI memo 1180, MIT, Dec. 1989.

No context found.

Maes, P. How To Do The Right Thing, AI-Laboratory, Vrije Universiteit Brussel and AI-Laboratory MIT, 1989.

No context found.

P. Maes, "How to do the right thing," Connection Science, vol. 1, no.3, 1989.

No context found.

Maes, P., "How to Do the Right Thing," Connection Science Journal, 1(3), pp. 291-323, 1989.

No context found.

Maes, P.: How to Do the Right Thing. Connection Science Journal, Vo. 1, 1989, No. 3, pp. 291--323.

No context found.

P. Maes. How to do the right thing. Connection Science Journal, 1:291--323, 1989.

No context found.

Maes. P. How to Do the Right Thing. Connection Science, Vo. 1, No. 3, p.p. 291-323,1989.

No context found.

P. Maes. How to do the right thing. Connection Science Journal, 1:291--323, Dec 1989.

First 50 documents  Next 50

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