Enhancing Cognitive System Self Perception Object Identification Using Neural Network

Size: px

Start display at page:

Download "Enhancing Cognitive System Self Perception Object Identification Using Neural Network"

Clara Austin
5 years ago
Views:

Technology, Volume-2, Issue-3, May-June, 2014, pp. 44-49, IASTER 2014 www.iaster.

Science, Mizan Tepi University, Tepi, Postal Box-121, Ethiopia 2 Department of Information Technology, Mizan Tepi University, Tepi, Postal Box-121, Ethiopia ABSTRACT This research paper deals with an

1 Technology, Volume-2, Issue-3, May-June, 2014, pp , IASTER Online: , Print: Enhancing Cognitive System Self Perception Object Identification Using Neural Network Saroj Kumar Gupta 1, Haftom G Egziabher 2 1 Department of Computer Science, Mizan Tepi University, Tepi, Postal Box-121, Ethiopia 2 Department of Information Technology, Mizan Tepi University, Tepi, Postal Box-121, Ethiopia ABSTRACT This research paper deals with an idea of object perception with object identification for cognitive system robotics. Initiative behind this agent development is to percept the object and percept human action and performs the same action with object identities. Object identities and action driven development agent is difficult for identifying whole body movement and challenging to identify pattern of whole object. We are working and trying to implement some background knowledge into robot which act and identify after perceiving object and its action. Cognitive Robot percept small perception and it perform complex action from it. This proposed robot achieves object identification, learning of environment and action perception. Keywords: Action Perception, Mirror Neuron, Utility theory, Neural Network. I. INTRODUCTION HOW a robot percept the identify object and percept human actions? Some mammal like rhesus macaques is able to identify their own body change of genes even when they are technically modify based on T cell and presence and absence of methyl based on and off genes, same way robot identify these kinds of cognitive functions may have the potential to break the limits of hand coded machine intelligence. The goal of this research paper is to create a cognitive system robot which actively develops perception, identifies object and absorb self and actions in non stationary environments. Our claim for current cognitive systems is that robot actions are perception developed with BG information, but their perception is not adapted as to give identity of object. In fact action driven development of perceptual ability is missing in robot learning in non stationary environments to identify object behavior. Therefore, identifying object and action perception of human in robots is not yet reconstrucable and perception of actions does by robots and humans is not treated in the same way at a perceptual level of robot. Artificial intelligence, makes it possible to design robot which can identify all existing specific elements object[1]. Fig 1. Rhesus Macaques and Neural Bio-Model Mirror Perception. The AI robot or computer gathers facts about a situation through sensors or human input. The computer compares this information to stored data and decides what the information signifies. The computer runs through various possible actions and predicts which action will be most successful based on the collected information and have memory space to keep track of object identification with the help of background knowledge (BG). 44

1.1. Humanoid Robot Humanoid Robotics includes a rich diversity of projects where perception, processing and action are embodied in a recognizably anthropomorphic form in order to emulate some subset

2 1.1. Humanoid Robot Humanoid Robotics includes a rich diversity of projects where perception, processing and action are embodied in a recognizably anthropomorphic form in order to emulate some subset of the physical, cognitive and social dimensions of the human body and experience. Figure 1 shows Schematic presentation of action-driven developments. An agent generates an action and associates the perceived the sensory event. The casual relation constructs body definition, motor control and action perception. The goal is not, to make robot that can be mistaken for or used interchangeably with real human beings. Rather, the goal is to create a new kind of tool and algorithm which, fundamentally different from any we have yet seen because it is designed to work with humans as well as to identify object. Humanoids will interact socially with people in typical, everyday environments. We already have robots to do tedious, repetitive labor for specialized environments and tasks. Instead, humanoids will be designed to act safely alongside humans, extending our capabilities in a wide variety of tasks and environments. II. ACTION PERCEPTION To perceive, according to this enactive approach to perception, is not merely to have sensations; it is to have sensations that we understand. In Action in Perception, we investigate the forms this understanding can take. He begins by arguing, on both phenomenological and empirical grounds, that the content of perception is not like the content of a picture; the world is not given to consciousness all at once but is gained gradually by active inquiry and exploration. we then argues that perceptual experience acquires content thanks to our possession and exercise of practical bodily knowledge, and examines, among other topics, the problems posed by spatial content and the experience of color. We consider the perspective aspect of the representational content of experience and assess the place of thought and understanding in experience. Finally, we explore the idea of the enactive approach for our understanding of the neuroscience of perception. An agent generates an action and associates the perceived the sensory event. The casual relation constructs body definition, motor control and action perception. Fig. 2. Schematic Presentation of Action-Driven Developments 2.1 Visual Motion A robot generates an motor exploration with the arm motor synergy. The synergy in this paper means coordination in the movement of multiple joint motors, based on motor exploration, the robot identifies its own body using vision and proprioception. We use visual motion cues to segment the robot s body parted from the background, since visual motion cues prove the target s independent from environment The absolute subtraction between the successive frames of monochrome image I m (x,t) result in a different image I f (x,t) as given I f (x,t) = I m (x,t)- I m (x,t-τ) (1) Where x=(ξ,ɳ ) denotes the horizontal and vertical coordinate on the image. T and τ denotes the sampling time and the interval of the frames [2]. 45

The disk of the i th point x i is represented as follows: D i (x) = { x x - x i r i } (2) Where r i denotes the radius of the disk.

3 We will now define a procedure for clustering different blobs and filling in the area. First, motion points/pixels are grouped in clusters. A set of points is randomly sampled from the high intensity points on I f. Each sample point is given a small disk. The disk of the i th point x i is represented as follows: D i (x) = { x x - x i r i } (2) Where r i denotes the radius of the disk. The neighbor disks are grouped as a new disk, if the disks intersect. The intersection of disk D i (x) and D j (x) x i - x j < r i - r j (3) 2.2 Body Identification Fig. 3. Visual Motion Detection We will now introduce the body identification procedure that allows a robot to segment its body from the environment. The assumption is that the causal relation between a self generated action and its effect define the body of the agent. The robot monitors the visuomotor correlation between proprioceptive and visual motion. When the robot detects the visuomotor Correlation, the visually moving object is identified as a part of the body. We have improved the single body part identification of the previous system to allow multiple body part identification as follows: the robot generates actions with each motor unit (e.g., the wrist or shoulder of the left or right side), and associates the sensory event with the actuated motor unit individually. Multiple body part identification enables the robot to perceive its own body parts and link them to the corresponding motor units. The robot performs repetitive movements to exclude other objects from body identification. Fig. 4. Motor Control of Robot Figure 4 illustrates this procedure. The advantage of this technique is that the actiondriven perception generalizes the body identification in which the body can be modified or extended by a grasped tool as shown in figure 4 which shows how to control body by perception from the environment. The robot generates a movement which is given u = q + δq (4) Where u denotes the motor command of the motor unit, q denotes the reference encoder values of the motor unit, and δq denotes a variation. 46

4 III. MIRROR NEURON SYSTEM Research has been conducted to locate where in the brain specific parts and neurological systems are activated when humans imitate behaviors and actions of others, discovering a mirror neuron system. This neuron system allows a person to observe and then recreate the actions of others. Mirror neurons are pre motor and parietal cells in the macaque brain that fire when the animal performs a goal directed action and when it sees others performing the same action. Evidence suggests that the mirror neuron system also allows people to comprehend and understand the intentions and emotions of others. Problems of the mirror neuron system may be correlated with the social inadequacies of autism. There have been many studies done showing that children with autism, compared with typically developing children, demonstrate reduced activity in the frontal mirror neuron system area when observing or imitating facial emotional expressions[5][7] self survey response shown with survey about percept information and algorithm simple problem solving and knowledge base agent solve the problem of perception of robot. 3.1 Action Perception Proposed action perception system is summarized as follows: (1) The action perception system is developed by observing the robot s self-generated actions. (2) The motor repertoire is constructed incrementally by combining learned primitives. (3) The sensory effect of an action is encoded in multimodal sensory space. (4) Human actions are predicatively recognized via intermediate evaluation of the sensory effect. (5) Action perception allows cross-modal sensory anticipation and action reproduction. 3.2 Problem Formulation Process of deciding what actions and states to consider, and follows goal formulation. A Simple Problem-Solving Agent Function SIMPLE-PROBLEM-SOLVING-AGENT(P) returns an action inputs: p, a percept static: s, an action sequence, initially empty state, some description of the current world state g, a goal, initially null problem, a problem formulation state < UPDATE-STATE(state, p) if s is empty then g FORMULATE-GOAL(state) problem < FORMULATE-PROBLEM(state, g) s SEARCH( problem) action RECOMMENDATION^, state) s < REMAINDER(s, state) return action 3.3. A Generic Knowledge-Based Agent function KB-AGEN1( percept) returns an action static: KB, a knowledge base t, a counter, initially 0, indicating time TELL(KB, MAKE-PERCEPT-SENTENCE(percept, t)) action ASK(KB, MAKE-ACTION-QUERY(t)) TELL(KB, MAKE-ACTION-SENTENCE(action, t)) T t+ 1 return action 47

5 IV. BASIC UTILITY THEORY The principle of Maximum Expected Utility (MEU) seems like a reasonable way to make decisions, but it is by no means obvious that it is the only rational way. After all, why should maximizing the average utility be so special why not try to maximize the sum of the cubes of the possible utilities, or try to minimize the worst possible loss? Also, couldn't an agent act rationally just by expressing preferences between states without giving them numeric values? Finally, why should a utility function with the required properties exist at all? Perhaps a rational agent can have a preference structure that is too complex to be captured by something as simple as a single real number for each state. Constraints on rational preferences these questions can be answered by writing down some constraints on the preferences that a rational agent should have, and then showing that the MEU principle can be derived from the constraints. Writing down these constraints is a way of defining the semantics of preferences. The idea is that, given some preferences on individual atomic states, the theory should allow one to derive results about preferences for complex decision-making scenarios. This is analogous to the way that the truth value of a complex logical sentence is derived from the truth value of its component propositions, and the way the probability of a complex event is derived from the probability of atomic events. In the language of utility theory, the complex scenarios are called lotteries to emphasize the idea that the different attainable outcomes are like different prizes, and that the outcome is determined by chance. The lottery L, in which there are two possible outcomes state A with probability) and state B with the remaining probability is written as L=[p,A; 1-p,B] In general, a lottery can have any number of outcomes. The following notation is used to express preferences or the lack of a preference between lotteries or states: AB- A is preferred to B., AB- the agent is indifferent between A and B. AB- the agent prefers A to B or is indifferent between them. The following six constraints are known as the axioms of utility theory. They specify the most obvious semantic constraints on preferences and lotteries. Ordability: Given any two states, a rational agent must either prefer one to the other or else rate the two as equally preferable. That is, an agent should know what it wants. (A B) V (B A) V (A ~ 5) (5) Transitivity: Given any three states, if an agent prefers A to B and prefers B to C, then the agent must prefer A to C. (A B) A (BC) => (A C) (6) Continuity: If some state B is between A and C in preference, then there is some probability p for which the rational agent will be indifferent between getting B for sure and the lottery that yields A with probability/? and C with probability 1 p. A B y C => p [p, A; 1 - p, C] ~ B (7) Substitutability: If an agent is indifferent between two lotteries, A and B, then the agent is indifferent between two more complex lotteries that are the same except that B is substituted for A in one of them. This holds regardless of the probabilities and the other outcome(s) in the lotteries. A ~ B => [p,a; l-p,c]~ [p,b;l -p,c] (8) 48

6 Monotonicity: Suppose there are two lotteries that have the same two outcomes, A and B. If an agent prefers A to B, then the agent must prefer the lottery that has a higher probability for A (and vice versa). AB => (pq [p,a; l-p,b] [q,a; 1 -q,b]) (9) Dccomposability: Compound lotteries can be reduced to simpler ones using the laws of probability. This has been called the "no fun in gambling" rule because it says that an agent should not prefer (or disprefer) one lottery just because it has more choice points than another. [p, A; 1 p, [q,b; 1 q,c]\ ~ [p,a; (1 p)q,b; (1 p)(l q),c] (10) Suppose that an agent is situated in the environment shown below Beginning in the start state, it must execute a sequence of actions. The environment terminates when the agent reaches one of the states marked +1 or -1. In each location, the available actions are called North, South, East, and West. We will assume for now that the agent knows which state it is in initially, and that it knows the effects of all of its actions on the state of the world [8]. V. CONCLUSION We propose a robot as an agent based on human action perception and experimentally shown as with clapping sound robot. This works as self generated action and perception through sensor motor. An ability of lacking tools and hardware component tools are hard to get to work with motor control but still working to achieve it. REFERENCES [1] A. Iriki, M. Tanaka, and Y. Iwamura, Coding of Modified Body Schema During Tool Use by Macaque Postcentral Neurones, Neuroreport, Vol. 7, No. 14, pp , [2] A. Iriki, M. Tanaka, S. Obayashi, and Y. Iwamura, Self-images in the Video Monitor Coded by Monkey Intraparietal Neurons, Neurosci. Res., Vol. 40, No. 2, pp , [3] G. Rizzolatti, L. Fadiga, V. Gallese, and L. Fogassi, Premotor Cortex and the Recognition of Motor Actions, Cognit. Brain Res., Vol. 3, No. 2, pp , [4] V. Gallese, L. Fadiga, L. Fogassi, and G. Rizzolatti, Action recognition in the premotor cortex, Brain, Vol. 119, No. 2, pp , Oct [5] L. Fogassi, P. Ferrari, B. Gesierich, S. Rozzi, F. Chersi, and G. Rizzolatti, Parietal lobe: From Action Organization to Intention Understanding, Science, Vol. 308, No. 5722, pp , 2005 [6] [7] perception of self action by Ryo Saegusa [8] Artificial Intelligence a Modern Approach By J. Russel and Peter Norvig. [9] ectsgenes 49

EMBODYING GESTURE PROCESSING A Computational Cognitive Model for Humanoid Virtual Agents

EMBODYING GESTURE PROCESSING A Computational Cognitive Model for Humanoid Virtual Agents Amir Sadeghipour PhD Student @ Sociable Agents Group Bielefeld University, Germany Supervisor: Prof. Stefan Kopp