Method and device for the control of an autonomously exploring robot
First Claim
Patent Images
1. A method for controlling an autonomously exploring robot comprising the steps of:
- capturing values, using a sensory device, which are fed as input values into an artificial neuronal strategy generating network,generating control values with the strategy generating network, according to input and trained synapse strengths, which control acting units of the robot,feeding the values captured by the sensory device and the control values generated by the strategy generating network into an artificial neuronal network system including a plurality of further networks,storing said input values of the sensory device and corresponding control values, which constitute pairs of values, into a short term memory,distinguishing between pairs of values learned during a training phase and unknown pairs of values with a dynamic analyzer on a criterion of dynamic stability of the network system,extending an initial repertoire of said strategy generating network and the network system by assimilation of selected values captured by the sensory device and the control values generated with the strategy generating network in a training process when the dynamic analyzer identifies a pair of values as known and belonging to a successful string of values,selecting unknown pairs of stored values which precede the known pair of values,training the network system until adaptation of synapses strengths of the network system results in a stable output pattern,feeding the strategy generating network with the stable output pattern, and the selected unknown pairs of values, andtraining the strategy generating network, until all selected pairs of stored values are trained, thereby extending the initial repertoire of the strategy generating network and network system.
2 Assignments
0 Petitions
Accused Products
Abstract
In a control method and a control device for an autonomously exploring robot, sensory values induce, in a strategy generating network, control sets according to inner connectivity which is acquired in a training phase. The control sets control acting units which act on the environment. Control sets and sensory values are given as input to a network system which distinguishes between already trained pairs of values and unknown pairs of values.
-
Citations
33 Claims
-
1. A method for controlling an autonomously exploring robot comprising the steps of:
-
capturing values, using a sensory device, which are fed as input values into an artificial neuronal strategy generating network, generating control values with the strategy generating network, according to input and trained synapse strengths, which control acting units of the robot, feeding the values captured by the sensory device and the control values generated by the strategy generating network into an artificial neuronal network system including a plurality of further networks, storing said input values of the sensory device and corresponding control values, which constitute pairs of values, into a short term memory, distinguishing between pairs of values learned during a training phase and unknown pairs of values with a dynamic analyzer on a criterion of dynamic stability of the network system, extending an initial repertoire of said strategy generating network and the network system by assimilation of selected values captured by the sensory device and the control values generated with the strategy generating network in a training process when the dynamic analyzer identifies a pair of values as known and belonging to a successful string of values, selecting unknown pairs of stored values which precede the known pair of values, training the network system until adaptation of synapses strengths of the network system results in a stable output pattern, feeding the strategy generating network with the stable output pattern, and the selected unknown pairs of values, and training the strategy generating network, until all selected pairs of stored values are trained, thereby extending the initial repertoire of the strategy generating network and network system. - View Dependent Claims (2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24)
-
-
25. A device for controlling an autonomously exploring robot comprising:
-
an artificial neuronal strategy generating network which generates control values for acting units of the robot according to input and trained synapse strengths, a sensory device for capturing values from an environment which are fed into the strategy generating network as input values, acting units which receive control values from the strategy generating network, a short term memory for storing said input values of the sensory device and corresponding control values, which constitute a pair of values, an artificial neuronal network system, including a plurality of further networks, which is supplied with the values captured by the sensory device and the control values generated by the strategy generating network, generates stable output values upon receiving a pair of values already learned during a training process, and generates instable output values upon receiving an unknown pair of values, output values of the artificial neuronal network system being supplied to a signal generating network, and a dynamic analyzer for distinguishing between pairs of values learned during a training phase and unknown pairs of values on the criterion of dynamic stability of the network system, an output of which may be connected to the strategy generating network. - View Dependent Claims (26, 27, 28, 29, 30, 31, 32, 33)
-
Specification