James S. Albus. Brains, Behavior, and Robotics. BYTE Books, 1981.  Home/Search   Document Not in Database   Summary   Related Articles   Check

....its taxonomy see [8] One very useful tool when fighting the computational complexity of planning is the creation of hierarchies of planners. The Real time Control System (RCS) reference model architecture is one such architecture and it has been successfully applied to multiple diverse systems [1, 3]. The target systems for RCS are in general, complex control problems. Although it has been shown [2, 10] that the complexity of a control problem is reduced by the use of a hierarchical control system, the reduction of error as a function of complexity at one level of the hierarchy has been ....

J.S. Albus. Brain, Behavior, and Robotics. McGraw-Hill, 1981.

....level (slower replanning at the top of the hierarchy and faster replanning at the bottom of the hierarchy) 4. Distribution of computational burden between levels (i.e. distributed computing) 5. Efficient compression of representation. In our case, we follow the RCS hierarchical archi tecture [3], 4] RCS is a hierarchical control system divided into a multiplicity of levels of resolutions. At the top of the hierarchy, the representation is coarse with large time and space horizons and slow replanning cycles. At the bottom of the hierarchy the opposite is true. Fast replanning cycles at ....

J.S. Albus, Brain, Behavior, and Robotics, McGraw-Hill, 1981.

....size of the region, the number of keepers, and the number of takers. A sample starting con guration with 4 keepers and 3 takers on a 30m 30m eld is shown in Figure 1. In their previous work, Stone Sutton [16] apply episodic SMDP Sarsa( with linear tile coding function approximation (CMACs [1]) to the keepaway task. The learners choose not from the simulator s primitive actions (e.g. kick, dash, and turn) but from higher level actions constructed from the following set of basic skills (implemented by the CMUnited 99 team [15] HoldBall( Remain stationary while keeping possession of ....

J. S. Albus. Brains, Behavior, and Robotics. Byte Books, Peterborough, NH, 1981.

.... 3 REINFORCEMENT LEARNING AND FUNCTION APPROXIMATION We used a simple CMAC network to represent the approximate value function, as described by Sutton (1996) CMAC networks require less memory than table lookup approaches and possess excellent convergence speed and solution quality (e.g. see Albus, 1981; Miller et al. 1990) We have also obtained satisfactory results with backpropagation networks, but they always learn much slower than CMAC networks, particularly in the initial stages of training. In this paper s experiments we used a CMAC consisting of 10 tilings, each a simple 10 by 10 grid. ....

Albus, J. S. (1981). Brain, Behavior, and Robotics, chapter 6, pages 139--179. Byte Books.

....execute PassBall(f) Else execute HoldBall( 4 Function Approximation In large state spaces, RL relies on function approximation to generalize across the state space. To apply RL to the keepaway task, we used a sparse, coarse coded kind of function approximation known as tile coding or CMAC [1, 8, 20, 6, 4, 17]. This approach uses multiple overlapping tilings of the state space to produce a feature representation for a nal linear mapping where all the learning takes place (see Figure 3) Each tile corresponds to a binary feature that is either 1 (the state is within this tile) or zero (the state is not ....

J. S. Albus. Brains, Behavior, and Robotics. Byte Books, Peterborough, NH, 1981.

....The trial starts at the bottom of the mountain with x 0 = Gamma0:5 and v 0 = 0. Reward is 1 on all time steps. The trial terminates with the first position value that exceeds x t 1 0:7. We used a simple CMAC network to represent the approximate value function, as described by (Sutton 1996) (Albus 1981). CMAC networks require less memory than table lookup approaches and possess excellent convergence speed and solution quality, e.g. see (Miller, Glanz, Kraft 1990) We have also obtained satisfactory results with backpropagation networks, but they always learn much slower than CMAC networks, ....

Albus, J. 1981. Brain, Behavior, and Robotics. Byte Books.

....[7] execute PassBall(f) Else execute HoldBall( 4 Function Approximation In large state spaces, RL relies on function approximation to generalize across the state space. To apply RL to the keepaway task, we used a sparse, coarsecoded kind of function approximation known as tile coding or CMAC [1, 16]. This approach uses multiple overlapping tilings of the state space to produce a feature representation for a nal linear mapping where all the learning takes place (see Figure 3) Each tile corresponds to a binary feature that is either 1 (the state is within this tile) or zero (the state is not ....

J. S. Albus. Brains, Behavior, and Robotics. Byte Books, Peterborough, NH, 1981.

....with the first position value that exceeds x t 1 0:5. We used a simple CMAC network to represent the approximate value function, as described by (Sutton 1996) CMAC networks require less memory than table lookup approaches and possess excellent convergence speed and solution quality, e.g. see (Albus 1981), Miller, Glanz, Kraft 1990) We have also obtained satisfactory results with backpropagation networks, but they always learn much slower than CMAC networks, particularly in the initial stages of training. In this paper s experiments we used a CMAC consisting of 5 tilings, each a simple 9 by 9 ....

Albus, J. 1981. Brain, Behavior, and Robotics. Byte Books.

....architecture in which intermediate levels exhibit intermediate values on each of the dimensions of difference indicated above. It is very much influenced by the work of Albus concerning the organization of the human nervous system and its implications for the design of complex computer agents [Albus, 1981; 1992; see also: Brooks, 1986; Firby, 1992; Gat, 1992; Hayes Roth, 1990; Hayes Roth, Erman, and Hayes Roth, 1992] Although, at first glance, the two level structure may resemble several two level planner reactor architectures [Drummond, Swanson, Bresina, and Levinson, 1993; Bresina and Drummond, ....

Albus, J.S. Brains, behavior, and robotics. Peterborough, N.H. : BYTE Books, 1981.

....to appear in the popular press and agents are becoming a recognizable concept to the lay public. Table 1. Intelligent Agents . Problem solving agents Hayes Roth, 1985; Newell, 1990 . Personal assistants Etzioni Weld, 1994; Kautz etal, 1994; Maes, 194; Mitchell etal, 1994 . Robotic agents Albus, 1981; Arkin, 1985; Brooks, 1986; Firby, 1992; Gat, 1990; McDermott, 1992 Hayes Roth, etal, 1995; Laird et al., 1989; Vere Bickmore, 1990 . Agents for toy problems Agre Chapman, 1987; Bresina Drummond, 1990; Pollack, 1991 . Monitoring agents Hayes Roth, 1989 Table 2. Research Issues for ....

Albus, J.S. Brains, behavior, and robotics. Peterborough, N.H. : BYTE Books, 1981.

....a new component of the subgoal graph 2 . Connected random configurations are possibly inserted as candidates to a subgoal following the chosen strategy 3 . Finally we try to merge all components. 3. Speedup Using Neural Preprocessing For neural preprocessing we use a slightly modified CMAC [1,6], which is a general, locally generalizing function approximator. CMAC should map each configuration c 2 C free in configuration space on to a connected subgoal sg 2 SG with small Euclidean distance. The information about the incrementally growing set of subgoals should also be available to ....

Albus, J. S.: Brains, Behaviour, and Robotics. Byte Books (1981)

....[137, 25] This process of decomposition can be repeated for subtasks at each succeeding level until a subtask is decomposed into a sequence of elemental control actions. Such a decomposition enables us to represent any task as a trajectory of subtasks at each level of decomposition (see Albus [3]) A set of tasks is defined to be decomposable if the set can be divided into at least two disjoint subsets of tasks of different levels of difficulty, such that (1) any task of a level of difficulty can be decomposed into a sequence of tasks of lower levels of difficulty, and (2) the tasks at ....

Albus, J. S. Brains, behavior, and Robotics. BYTE Books, Peterborough, NH, 1981.

....sensor data is unknown to the agent, the only task it faces is to generalize the relationship between its inputs (horizontal position, vertical position and predator indication) and the success value it receives. This learning task could be solved by any current generalization method (e.g. CMAC [Albus, 1981], radial basis functions, neural networks) and could even be learned without generalization by plain table lookup, although the number of training examples required in that case would be unnecessarily high, since each predator would have to be encountered in combination with each position of ....

Albus, J. (1981). Brains, Behavior, and Robotics. Byte Books, Peterborough, NH.

....above and, more generally, higher levels would organize computations at higher 6 levels of abstraction. Our current architecture has only two levels because our current applications do not require finer resolution along these dimensions. Many researchers present similar leveled architectures [Albus, 1981; 1991; Connell, 1992; Firby, 1992; Gat, 1992; Hayes Roth, 1990; Hayes Roth, Erman, and HayesRoth, 1992, Terry, 1994] Although, at first glance, the two level structure may resemble several two level planner reactor architectures [Drummond, Swanson, Bresina, and Levinson, 1993; Bresina and ....

Albus, J.S. Brains, behavior, and robotics. Peterborough, N.H. : BYTE Books, 1981.

.... of memories derives its associative and robust nature by sparsely sampling the input space and distributing the data among many independent agents (Kanerva, 1992) Other members of this class include a model of the cerebellar cortex (Marr, 1969) the Cerebellar Model Arithmetic Computer (CMAC) (Albus, 1981), and Sparse Distributed Memory (SDM) Kanerva, 1988) First, we present a simplified account of the immune response and immunological memory. Next, we present SDM, and then we show the correlations between immunological memory and SDM. Finally, we show how associative recall in the immune ....

Albus, J. S. (1981). Brains, Behavior, and Robotics. Byte Books, Peterborough, NH.

....adaptation the robots can achieve based on their finite set of primary behaviours. Keywords Autonomous Robots, Adaptive, Fuzzy Logic. 1. Introduction The field of autonomous robots has advanced significantly over the past decade, since the advent of reactive or bottom up methods advocated by [1,8,3]. Recent developments however, have now emphasised a hybrid approach, which aims to integrate the task planning capacity of traditional A.I based architectures with reactive control methods, 12,25] This approach appears to offer the best of both philosophies, in providing an agent with strategic ....

Albus J.S., Brains, Behaviour, and Robotics, Byte Books, McGraw-Hill, Peterborough, 1981.

....the type of manipulation tasks routinely performed by biological organisms, it seems that the approach of controlling robotic manipulator systems by a mathematical formalism such as trigonometric equations is inadequate to produce truly sophisticated motor behavior. To remedy this situation, Albus [1,2, 3] proposed a memory driven, table reference motor control system called Cerebellar Model Articulation Controller (CMAC ) The fact that for n input variables with R distinguishable levels there are R n possible inputs may be sufficient to discourage this line of research. However, Albus observed ....

J. S. Albus, Brains, Behaviour, and Robotics, Byte Books, Peterborough, NH, 1981.

....the probability of selecting action a in state x is equal Prob(x; a ) exp(Q(x; a ) T ) P a exp(Q(x; a) T ) 11) where the temperature T 0 adjusts the amount of randomness. 3 CMAC: A Coarse Coded Look Up Table CMAC (Cerebellar Model Articulation Controller) was proposed by Albus [ 1981 ] partly as a model of the functionality of the cerebellum, and partly as a useful method for function storage and approximation. Its advantages had not been very widely known until the end of the eighties; since then it has been more and more often successfully applied, mainly in automatic ....

.... had not been very widely known until the end of the eighties; since then it has been more and more often successfully applied, mainly in automatic control, e.g. Miller et al. 1990 ] The particular implementation of CMAC used in this paper strictly follows the original description of Albus [ 1981 ] To get a rough idea of how it works, we may think of it as of a sparse coarse coded look up table which uses multiple overlapping tilings to cover the input space, so that each input point is contained in a number of overlapping tiles. The tilings are displaced relative to each other by a ....

J. S. Albus. Brain, Behavior, and Robotics. BYTE Books, 1981.

....system will be presented. 1 Introduction A long standing goal has been to create a reference model architecture for intelligent controllers. The Real time Control System (RCS) reference model architecture is one such architecture and it has been applied successfully to multiple diverse systems [1, 2]. The target systems for RCS are, in general, complex control problems. It has been shown [3] that the complexity of a control problem is reduced by the use of a hierarchical control system. In order to take advantage of this fact, RCS provides guidelines for the decomposition of the problem into ....

J.S. Albus. Brain, Behavior, and Robotics. McGraw-Hill, 1981.

No context found.

James S. Albus. Brains, Behavior, and Robotics. BYTE Books, 1981.

No context found.

J.S. Albus, Brain, Behaviour, and Robotics, Byte Books, Peterborough, NH, 1981.

No context found.

J. S. Albus. Brains, Behavior, and Robotics. Byte Books, Peterborough, NH, 1981.

No context found.

J. S. Albus. Brains, Behavior, and Robotics. Byte Books, Peterborough, NH, 1981.

No context found.

J. Albus. Brains, Behaviors, and Robotics. BYTE Books, 1981.

No context found.

J.S. Albus. Brains, Behavior, and Robotics . Peterborough, N.H. : BYTE Books, 1981.

First 50 documents

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