Device and method for ascertaining a physical property of a physical object

A method for controlling an actuator, comprising the following steps:

• ascertaining a physical property of a physical object by:
• detecting sensor data using an event-based sensor for each input point in time of a sequence of input points in time, the sensor data including pieces of information about the physical object;
• for each subsequence of a breakdown of the sequence of input points in time into multiple subsequences:
• feeding the sensor data detected for the input points in time of the subsequence to a pulsed neural network,
• processing the sensor data of the input points in time of the subsequence by pulsed neurons of the pulsed neural network for generating a first processing result of the subsequence, the pulsed neurons in each case integrating values of sensor data of different input points in time of the subsequence;
• feeding the first processing result of the subsequence to a non-pulsed neural network, and
• processing the first processing result of the subsequence by non-pulsed neurons of one or multiple first layer(s) of the non-pulsed neural network for generating a second processing result of the subsequence;
• feeding the second processing results of the multiple subsequences to one or multiple second layer(s) of the non-pulsed neural network;
• combining the second processing results of the multiple subsequences by non-pulsed neurons of the one or multiple second layer(s) of the non-pulsed neural network in that the non-pulsed neurons of the one or multiple second layer(s) in each case calculate a weighted sum of values of second processing results of different subsequences;
• and
• ascertaining the physical property of the physical object from outputs of the non-pulsed neurons of the one or multiple second layer(s);
• and
• controlling an actuator as a function of the ascertained physical property of the physical object.

A method for detecting an anomaly of a physical object, comprising the following steps:

• ascertaining a physical property of a physical object by:
• detecting sensor data using an event-based sensor for each input point in time of a sequence of input points in time, the sensor data including pieces of information about the physical object;
• for each subsequence of a breakdown of the sequence of input points in time into multiple subsequences:
• feeding the sensor data detected for the input points in time of the subsequence to a pulsed neural network,
• processing the sensor data of the input points in time of the subsequence by pulsed neurons of the pulsed neural network for generating a first processing result of the subsequence, the pulsed neurons in each case integrating values of sensor data of different input points in time of the subsequence;
• feeding the first processing result of the subsequence to a non-pulsed neural network, and
• processing the first processing result of the subsequence by non-pulsed neurons of one or multiple first layer(s) of the non-pulsed neural network for generating a second processing result of the subsequence;
• feeding the second processing results of the multiple subsequences to one or multiple second layer(s) of the non-pulsed neural network;
• combining the second processing results of the multiple subsequences by non-pulsed neurons of the one or multiple second layer(s) of the non-pulsed neural network in that the non-pulsed neurons of the one or multiple second layer(s) in each case calculate a weighted sum of values of second processing results of different subsequences;
• and
• ascertaining the physical property of the physical object from outputs of the non-pulsed neurons of the one or multiple second layer(s);
• comparing the ascertained physical property of the physical object to a reference property of the physical object;
• and
• establishing that the physical object has an anomaly when the ascertained physical property deviates from the reference property.

A device configured to ascertain a physical property of a physical object, the device configured to:

• detect sensor data using an event-based sensor for each input point in time of a sequence of input points in time, the sensor data including pieces of information about the physical object;
• for each subsequence of a breakdown of the sequence of input points in time into multiple subsequences:
• feed the sensor data detected for the input points in time of the subsequence to a pulsed neural network,
• process the sensor data of the input points in time of the subsequence by pulsed neurons of the pulsed neural network for generating a first processing result of the subsequence, the pulsed neurons in each case integrating values of sensor data of different input points in time of the subsequence;
• feed the first processing result of the subsequence to a non-pulsed neural network, and
• process the first processing result of the subsequence by non-pulsed neurons of one or multiple first layer(s) of the non-pulsed neural network for generating a second processing result of the subsequence;
• feed the second processing results of the multiple subsequences to one or multiple second layer(s) of the non-pulsed neural network;
• combine the second processing results of the multiple subsequences by non-pulsed neurons of the one or multiple second layer(s) of the non-pulsed neural network in that the non-pulsed neurons of the one or multiple second layer(s) in each case calculate a weighted sum of values of second processing results of different subsequences;
• and
• ascertain the physical property of the physical object from outputs of the non-pulsed neurons of the one or multiple second layer(s).

A non-transitory computer-readable memory medium on which are stored program instructions for ascertaining a physical property of a physical object, the program instructions, when executed by one or multiple processors, causing the one or multiple processors to perform the following steps:

• detecting sensor data using an event-based sensor for each input point in time of a sequence of input points in time, the sensor data including pieces of information about the physical object;
• for each subsequence of a breakdown of the sequence of input points in time into multiple subsequences:
• feeding the sensor data detected for the input points in time of the subsequence to a pulsed neural network,
• processing the sensor data of the input points in time of the subsequence by pulsed neurons of the pulsed neural network for generating a first processing result of the subsequence, the pulsed neurons in each case integrating values of sensor data of different input points in time of the subsequence;
• feeding the first processing result of the subsequence to a non-pulsed neural network, and
• processing the first processing result of the subsequence by non-pulsed neurons of one or multiple first layer(s) of the non-pulsed neural network for generating a second processing result of the subsequence;
• feeding the second processing results of the multiple subsequences to one or multiple second layer(s) of the non-pulsed neural network;
• combining the second processing results of the multiple subsequences by non-pulsed neurons of the one or multiple second layer(s) of the non-pulsed neural network in that the non-pulsed neurons of the one or multiple second layer(s) in each case calculate a weighted sum of values of second processing results of different subsequences;
• and
• ascertaining the physical property of the physical object from outputs of the non-pulsed neurons of the one or multiple second layer(s).