Adaptive autonomous agent with verbal learning

• US 5,802,506 A
• Filed: 05/26/1995
• Issued: 09/01/1998
• Est. Priority Date: 05/26/1995
• Status: Expired due to Fees

• Current Assignee: Hutchison, William
• Original Assignee: Hutchison, William
• Inventors: Hutchison, William
• Primary Examiner(s): Downs, Robert W.
• Application Number:
  US08/451,543
• Time in Patent Office: 1,194 Days
• Field of Search: 395/21, 395/22, 395/23, 395/24
• US Class Current: 706/20
• CPC Class Codes: G06N 3/04 Architecture, e.g. interconnection topology

Abstract

The invention is an autonomous adaptive agent which can learn verbal as well as nonverbal behavior. The primary object of the system is to optimize a primary value function over time through continuously learning how to behave in an environment (which may be physical or electronic). Inputs may include verbal advice or information from sources of varying reliability as well as direct or preprocessed environmental inputs. Desired agent behavior may include motor actions and verbal behavior which may constitute a system output and which may also function "internally" to guide external actions. A further aspect of the invention is an efficient "training" process by which the agent can be taught to utilize verbal advice and information along with environmental inputs.

Claims

1. An adaptive autonomous agent comprising:

a neural network having a plurality of input nodes for receiving input signals and a plurality of output nodes for driving output signals;

storage means for storing and accumulating at least two primary values wherein each of the accumulated primary values are coupled to one of the input nodes of the neural network; and

at least one primary value function generator coupled to receive the primary values from the storage means and generate a primary value signal.

2. An artificial adaptive agent having a plurality of input nodes for receiving input signals and a plurality of output nodes of generating output signals wherein the output signals are trained responses to the input signals, the agent comprising:

a primary value register for storing and accumulating one or more primary values;

means for adjusting the accumulated primary values based upon the responses each time the responses change, wherein the means for adjusting comprises;

means for determining a cost of each response with respect to each primary value; and

means for determining a benefit of each response with respect to each primary value;

means for decrementing the current primary value by the determined cost and incrementing the current primary value by the determined benefit each time a response is generated by the agent; and

sensors coupled to sense the accumulated primary values and having sensor outputs coupled to some of the input nodes of the agent.

3. An artificial adaptive agent having a plurality of input nodes for receiving input signals and a plurality of output nodes of generating output signals wherein the output signals are trained responses to the input signals, the agent comprising:

a primary value register for storing and accumulating one or more primary values;

means for adjusting the accumulated primary values based upon the responses each time the responses change;

sensors coupled to sense the accumulated primary values and having sensor outputs coupled to some of the input nodes of the agent;

mean s for combining a plurality of the output signals to determine a situation value variable;

means for calculating a change in the situation value variable as the output signals change;

means for combining the calculated change in the situation value variable with the calculated change in the primary value variable to determine a learning signal, wherein the learning signal is coupled to the network; and

means for monitoring the learning signal during training.

4. An adaptive autonomous agent with verbal learning comprising:

a first sensor for detecting the agents own response;

a second sensor for detecting a condition in an environment external to the agent;

a third sensor for detecting status of the agent;

an artificial neural network having at least one input node coupled to each first, second, and third sensor and having a plurality of network outputs;

a selector for coupling at least one of the network outputs to a selector output based upon a value of the network outputs, the selector also providing a situation value output;

at least one response actuator coupled to a selector output, wherein the response actuator is capable of affecting the condition in the environment external to the agent;

a primary value register storing one or more primary values;

means for adjusting each of the stored primary values;

a primary value function generator having inputs coupled to the primary value register and having an output, the primary value function generator serving to combine the stored primary values to generate a single primary value to signal the output;

a first difference generator coupled to the output of the primary value function generator and having first difference output indicating change in the primary value signal over time;

a second difference generator coupled to the situation value output and having a second difference output indicating change in the situation value over time;

a learning function generator coupled to the first and second difference outputs and generating a learning signal determined by combining the first and second difference output values; and

means controlled by the learning signal for changing connection weights within the artificial neural network.

5. The adaptive autonomous agent with verbal learning of claim 4 wherein the first sensor detects the agents own response by observing the at least one response actuator.

6. The adaptive autonomous agent with verbal learning of claim 4 wherein the first sensor detects the agent'"'"'s own response by observing the at least one response actuator.

7. The adaptive autonomous agent with verbal learning of claim 4 further comprising a plurality of recurrent nodes within the artificial neural network coupled to each input node.

8. The adaptive autonomous agent with verbal learning of claim 4 wherein the selector couples the network output having the maximum value to the selector output and the situation value is equal to the value of the highest value network output.

9. The adaptive autonomous agent with verbal learning of claim 4 further comprising at least two response actuators and at least two selector outputs, wherein the selector couples more than one network output to the selector output enabling parallel operation of the at least two response actuators.

10. The adaptive autonomous agent with verbal learning of claim 4 wherein the means for changing connection weights only changes weights that are currently activated.

11. The adaptive autonomous agent with verbal learning of claim 4 wherein the means for changing connection weights changes weights that have been activated within a predetermined number of previous time steps.

12. The adaptive autonomous agent with verbal learning of claim 4 wherein the means for adjusting the primary values is controlled by learning signal value.

13. An adaptive autonomous agent with verbal learning comprising:

an artificial neural network having a plurality of nodes and programmable-weight connections between the nodes;

a selector for coupling selected ones of the nodes to a selector output;

a situation value generator generating an output based upon the value of the nodes selected by the selector circuit;

a primary value generator generating an output based upon an accumulated value of the number of stored primary values;

a learning signal generator generating a learning signal based upon a computed change in the situation value and the accumulated primary value from one time step of the artificial neural network to a subsequent time step of the artificial neural network; and

means coupled to receive the learning signal for adjusting the stored primary values.