US Patent 11315421 Systems and methods for providing driving recommendations

Systems and methods for providing driving recommendations are disclosed herein. One embodiment receives, at an ego vehicle, first vehicle data and first encoded information from one or more other vehicles; constructs, from the first vehicle data, graph data representing how the ego vehicle and the one or more other vehicles are spatially related; inputs the graph data, the first vehicle data, second vehicle data pertaining to the ego vehicle, and the first encoded information to a graph convolutional network that outputs second encoded information; inputs the second encoded information and previously stored encoded information to a recurrent neural network that outputs a set of parameters to a mixture model; predicts acceleration of the one or more other vehicles using the mixture model; and generates a driving recommendation for the ego vehicle based, at least in part, on the predicted acceleration of the one or more other vehicles.