A Naturally Spurred Programming Design for a Wise Humanoid Robot.


77 views
Uploaded on:
Description
The procedures of mimicking so as to plan a wise robot may be encouraged the ... The robot can recognize its own disappointments and report them to the individual. ...
Transcripts
Slide 1

A Biologically Motivated Software Architecture for an Intelligent Humanoid Robot Richard Alan Peters II, D. Mitchell Wilkes, Daniel M. Gaines, and Kazuhiko Kawamura Center for Intelligent Systems Vanderbilt University Nashville, Tennessee, USA

Slide 2

Intelligence The capacity of a person to gain for a fact, to reason well, to recollect vital data, and to adapt to the requests of day by day living. (R. Sternberg, 1994). Insight has developed through advancement and is show in warm blooded animals. The procedures of outlining a savvy robot may be encouraged by copying the characteristic structures and elements of mammalian brains.

Slide 3

Topics Mammalian Brains ISAC, the Vanderbilt Humanoid Robot A Control System Architecture for ISAC A Partial Implementation Research Issues

Slide 4

The Structure of Mammalian Brains Krubitzer, Kass, Allman, Squire The development of structure Common components of neocortical association Species contrasts Memory and affiliation Attention Implications for robot control designs

Slide 5

The Evolution of Structure Figure: Leah Krubitzer

Slide 6

Common elements of Neocortical Organization Somatosensory Cortex (SI, SII) Motor Cortex (M) Visual Cortex (VI, VII) Auditory Cortex (AI) Association Cortex Size contrasts in cortical modules are unbalanced to size contrasts in cortex

Slide 7

Common elements of Neocortical Organization Figure: Leah Krubitzer

Slide 8

Species Differences Sizes and states of a particular cortical field Internal association of a cortical field Amount of cortex committed to a specific tactile or subjective capacity Number of cortical fields Addition of modules to cortical fields Connections between cortical fields

Slide 9

Memory: a Functional Taxonomy Squire Immediate memory : information supports for ebb and flow tangible data; holds data for around 0.1s Working memory : scratch-cushions, e.g. phono-consistent circle, visuospatial sketch cushion; the representation of tactile data in its nonappearance Short term memory (IM & WM) is a gathering of memory frameworks that work in parallel Long-term memory : can be reviewed for a considerable length of time; distinctive physically from STM

Slide 10

Memory: Biological Mechanisms Immediate memory — chemicals in neurotransmitter Working memory — increment in presynaptic vesicles; intra-neuron and between neuron protein discharge and transmitter movement Long-term memory — development of new neurotransmitters; requires translation of qualities in neurons.

Slide 11

Association The synchronous actuation of more than one tactile preparing range for a given arrangement of outside jolts A memory that connections numerous occasions or boosts Much of the neocortex not dedicated to tangible handling seems, by all accounts, to be included in affiliation

Slide 12

Memory and Sensory Data Bandwidth of signs out of tactile cortical fields is much littler than data transfer speed Sensory cortical fields all anticipate to territories inside affiliation cortex Suggests: Environment is examined for certain remarkable data, much is missed. Past recollections connected by affiliation fill in the crevices in data.

Slide 13

Attention: a Definition An evident grouping of spatio-transient occasions, to which a computational framework or subsystem designates a chain of command of assets. In that sense, the dynamic initiation of structures in the cerebrum is attentional .

Slide 14

Attention: Some Types Visual — where to look next Auditory — sudden onset or end of sound Haptic — found something Proprioceptic — entering insecure position Memory — activated by tactile data Task — activity choice Conscious — recallable occasion grouping

Slide 15

Attention: Executive Control Figure: Posner & Raichle

Slide 16

Mammalian Brains Have tangible handling modules that work ceaselessly in parallel Selectively channel approaching tactile information and supplement that data from memory through connection and affiliation Exhibit dynamic examples of action through nearby changes in cell digestion system — shifts in enactment

Slide 17

ISAC, a Two-Armed Humanoid Robot

Slide 18

Physical Structure of ISAC Arms: two 6 DOF impelled by pneumatic McKibben fake muscles Hands: human, pneumatic with closeness sensors and 6-hub FT sensors at wrists Vision: stereo shading PTV head Audition: client amplifier Infrared movement sensor cluster

Slide 19

ISAC Hardware under Construction Hybrid pneumatic/electric human hand Head mounted binaural mouthpiece framework Finger tip touch sensors

Slide 20

Computational Structure of ISAC Network of standard PCs Windows NT 4.0 OS Special equipment restricted to gadget controllers Designed under Vanderbilt\'s Intelligent Machine Architecture (IMA)

Slide 21

Low-Level Software Architecture: IMA Software operators (SA) plan model and devices SA = 1 component of an area level framework descr. SA firmly exemplifies all parts of a component SAs convey through message passing Enables simultaneous SA execution on independent machines on a system Facilitates between operators correspondence w/DCOM Can actualize any legitimate design

Slide 22

Primitive Agent Types Hardware : give deliberations of sensors and actuators, and low level handling and control (e.g., clamor sifting or servo-control circles). Conduct : e ncapsulate firmly coupled detecting - incitation circles. Could conceivably have runtime parameters . Environment: process sensor information to upgrade a reflection of nature. Can bolster practices, for example, ``move-to\'\' or ``fixate\'\' which require run-time parameters. Assignment : embody basic leadership capacities, and sequencing components for equipment, conduct, and environment operators.

Slide 23

Agent Object Model The specialists object model depicts how an operators system, characterized by the robot-environment model, is built from an accumulation of part questions

Slide 24

IMA Component Objects Agent Comp. — operators interfaces to chief and to diligent streams Policy Comp. — epitomizes an OS string Representation Comp. — a DCOM object that conveys an operators\' state to different specialists Mechanism Comp. — configurable articles that can be conjured to perform one of an arrangement of calculations Agent Links — interfaces characterized by representations Relationship Comp. — deal with an arrangement of operators connections to specifically redesign and/or utilize every member join

Slide 25

Properties of IMA Granularity — various sensible levels Composition — specialists can consolidate operators Reusable — can be consolidated for new functionalities Inherently Parallel — offbeat, simultaneous operation. Express Representation — sensor data is omnipresent Flat Connectivity — all operators are coherently equivalent w.r.t. tangible information access and actuator charges Incremental — All modules that deliver summons for the equipment work in an incremental mode

Slide 26

Correspondence of IMA Agents to System-Level Entities

Slide 27

A Bio-Inspired Control Architecture IMA can be utilized to actualize any control engineering. Singular operators can firmly couple detecting to incitation, and consolidate each other a la subsumption (Brooks). IMA Inter-specialists correspondences encourage engine constructions (Arkin). Structure of specialists which have level availability empowers half and half designs

Slide 28

ISAC Control System Architecture Primary Software Agents Sensory EgoSphere Attentional Networks Database Associative Memory Attentional Control through Activation Learning System Status Self Evaluation

Slide 29

Example Primary Software Agents: Visual consideration Color division Object acknowledgment Face acknowledgment Gesture acknowledgment Vision Audition Aural consideration Sound division Speech acknowledgment Speaker recognizable proof Sonic confinement

Slide 30

Example Primary Software Agents L & R Arm control L & R Hand control PTV movement Others Motor Infrared movement det. Peer operators Object specialists Attention specialists Sensory information recd\'s.

Slide 31

Higher Level Software Agents: Robot self specialists Human operators Object specialists (different) Visually guided getting a handle on Converse Reflex control Memory affiliation Visual following Visual servoing Move EF to FP Dual arm control Person Id Interpret V-Com Reflex control

Slide 32

Agents and Cortical Fields Agents can be intended to be practically similar to the regular field structure of the neocortex. Visual, sound-related, haptic, proprioceptic, attentional, and memory affiliation operators stay on continually and dependably change the current tangible inputs from the earth

Slide 33

Atlantis: a Three Layer Architecture Deliberator — world model, organizer Sequencer — undertaking line, agent, screens Controller — sensor/actuator coupled practices Erran Gat

Slide 34

Atlantis: General Schematic Figure: Erran Gat

Slide 35

Three-Layer Control with IMA Elements of control layer: specialists. Sequencing: through connections relying upon enactment vectors. (Because of level availability.) Deliberation: Various operators adjust the connections and actuation levels of others. (Because of composability.)

Slide 36

Virtual Three-Layer Architecture Deliberative Agent Sn IMA Agent S1 S3 ... S2 actuation join max initiation

Slide 37

3-Layer Control through Schemas Agents contend with each other for control of different specialists and the utilization of equipment assets. They do this by changing initiation vectors connected with specialists connections. In this way, the sequencer is proportionate to an engine composition

Slide 38

Handoff Task 1. Conjure Human Hand Env. Operators 2. Close Robot Gripper Res. Operators 3. Conjure Box Env. Specialists 4. Open Robot Gripper Move-to Move-to Box Env. Operators Open/Close Human Hand Env. Specialists Robot Gripper Res. Operators Activate Skin Color Tracking Beh. Specialists Visual Servoing Beh. Operators Primitive Agent Type Relationship Simple Task Agent Operation

Slide 39

Current Implementa

Recommended
View more...