Adaptive autonomous agent with verbal learning
First Claim
1. An adaptive autonomous agent comprising:
- an adaptive critic-type neural network having a plurality of network inputs at least some of which are verbal inputs applied in a selected sensory modality;
sensors connected to said network inputs;
a non-verbal network output;
actuators connected to said first network output;
means for sensing a current state of said actuators;
at least one verbal network output;
training means for establishing sets of connections from the verbal inputs to the first network output and the verbal network output.
0 Assignments
0 Petitions
Accused Products
Abstract
An autonomous adaptive agent (100) which can learn verbal as well as nonverbal behavior. The primary object of the system is to optimize a primary value function (7) over time through continuously learning how to behave in an environment (which may be physical or electronic). Inputs (1) may include verbal advice or information from sources of varying reliability as well as direct or preprocessed environmental inputs (1C). Desired agent (100) behavior may include motor actions and verbal behavior which may constitute a system output (3) and which may also function "internally" to guide external actions. A further aspect involves an efficient "training" process (306) by which the agent (100) can be taught to utilize verbal advice and information along with environmental inputs (1C).
29 Citations
14 Claims
-
1. An adaptive autonomous agent comprising:
-
an adaptive critic-type neural network having a plurality of network inputs at least some of which are verbal inputs applied in a selected sensory modality; sensors connected to said network inputs; a non-verbal network output; actuators connected to said first network output; means for sensing a current state of said actuators; at least one verbal network output; training means for establishing sets of connections from the verbal inputs to the first network output and the verbal network output. - View Dependent Claims (2, 3, 4, 12)
-
-
5. An adaptive agent configured to follow rules, each rule comprising an input vector paired with an output vector, the agent comprising:
-
a plurality of inputs responsive to events to provide input vectors to the adaptive agent; a plurality of outputs providing outputs to external components; a first repertoire comprising a plurality of members, each member comprising an abstract representation of an input vector; a second repertoire comprising a plurality of members, each member comprising an abstract representation of an output vector; a tact component operative to output an abstract representation from the first repertoire in response to a particular instance of an input vector; a pliance component operative to respond to an abstract representation from the second repertoire by outputting a particular instance of an output vector; and an intraverbal component operative to store associations between members of the first repertoire and members of the second repertoire. - View Dependent Claims (6, 7, 8, 10, 11)
-
-
9. An adaptive agent comprising:
-
a plurality of inputs responsive to events experienced by the agent to provide input vectors to the adaptive agent; at least one learning component responsive to the input vectors to learn a repertoire of mappings from the input vectors to output vectors; a rule following component coupled to at least some of the inputs, the rule following component operative to recognize a set of input vectors as a condition and create a condition-action pair comprising the recognized condition paired to a set of the learned output vectors, wherein the rule following component uses the learned repertoire to perform the pairing; wherein the at least one learning component further comprises; a primary value register for storing one or more primary values; means for adjusting the primary values based upon the responses each time the responses change; and sensors coupled to sense the primary values and having sensor outputs coupled to some of the input nodes.
-
-
13. An adaptive autonomous agent with verbal learning comprising:
-
an artificial neural network having verbal input nodes for receiving verbal information, verbal output nodes for expressing verbal information, and actuator output nodes for indicating an action wherein the agent responds to verbal information on the verbal input nodes by generating a signal on the actuator output nodes and generating a verbal response on the verbal output nodes such that the verbal response is a symbolic representation of the action indicated by the actuator output nodes; a plurality of non-verbal input nodes; a primary value register for storing one or more primary values; means for adjusting the primary values based upon the responses each time the responses change; and sensors coupled to sense the primary values and having sensor outputs coupled to some of the non-verbal input nodes.
-
-
14. A goal-directed agent comprising:
-
a neural network having a plurality of inputs and at least one output generating a response; a primary value register for storing one or more primary values, the primary value representing a goal; means for adjusting the primary values based upon the response each time the responses changes; and sensors coupled to sense the primary values and having sensor outputs coupled to some of the input nodes.
-
Specification