Computer-implemented system for controlling an autonomous vehicle and device for implementing a neural network system for training software embedded in an autonomous vehicle

Abstract

Computer-implemented system (100) for controlling an autonomous vehicle (10), aF (10), wherein the system (100) comprises: a non-volatile, computer-readable storage medium (32, 46) containing a set of instructions for programming at least one computer (34), wherein the at least one computer (34) comprises: a neural network (635, NN) with a large number of nodes with context-to-vector, context2vec, context-mediated embeddings to enable the operation of the autonomous vehicle (10); a variety of coded context2vec-aF words (610) in a sequence of time intervals to capture context and behavior data derived from autonomous vehicle sensor data (10) in time sequences, wherein the context and behavior data include: at least map, situation and behavior data of the autonomous vehicle operation (10); a set of inputs to the neural network (635, NN) which includes the following: at least one current, one previous and one subsequent coded context2vec-aF word (610), each represented in a one-hot scheme of a set of possibilities of the context2vec-aF words (610), wherein at least one context2vec-aF word (610) of the set is designated with an ON state, while other context2vec-aF words (610) of the set are designated with an OFF state; a neural network solution applied by the at least one computer (34) to determine a target context aF word of each set of inputs based on the current context2vec aF word (610); an output vector computed by the neural network (635, NN) representing the embedded distributed one-hot scheme of the input-encoded context2vec-aF word (610); and a set of behavior control operations for controlling the behavior of the autonomous vehicle (10) based on the context2vec-aF word (610) by the at least one computer (34); the map, situation and behavior data embedded in the context2vec-aF word (610) further include the following: a first, a second, a third, a fourth and a fifth part; the first part comprising the following: junction and stop map data; the second part includes the following: distance and time data until the collision; the third part includes the following: speed target and lane intention situation data; the fourth part comprises the following: data for forward clearance and expected maneuverability; and the fifth part includes the following: relative and expected maneuver behavior data to enable a control action by the at least one computer (34) of the autonomous vehicle (10).

Images