Predicting jaywalking behaviors of vulnerable road users

A system, comprising:

• a memory;
• and
• one or more computers configured to:
• obtain location data that identifies a current location of a vulnerable road user (VRU) detected within a vicinity of a vehicle traveling on a roadway;
• obtain environmental data that describes an environment of the VRU, wherein the environmental data identifies a plurality of environmental features in the environment of the VRU;
• determine a nominal heading of the VRU;
• generate a set of predictive inputs that indicate, for each of at least a subset of the plurality of environmental features, a physical relationship between the VRU and the environmental feature, wherein the physical relationship is determined with respect to the nominal heading of the VRU and the current location of the VRU;
• processing the set of predictive inputs with a heading estimation model to determine a predicted heading offset of the VRU, wherein the predicted heading offset indicates a difference between (i) a direction of a vector defined by the nominal heading of the VRU and (ii) a direction of a vector from the current location of the VRU to a predicted terminal location of the VRU's movements;
• and
• direct the vehicle to implement an autonomous driving decision generated using the predicted heading offset of the VRU.

One or more non-transitory computer-readable media encoded with instructions that, when executed by data processing apparatus, cause the data processing apparatus to perform operations comprising:

• obtaining location data that identifies a current location of a vulnerable road user (VRU) detected within a vicinity of a vehicle traveling on a roadway;
• obtaining environmental data that describes an environment of the VRU, wherein the environmental data identifies a plurality of environmental features in the environment of the VRU;
• determining a nominal heading of the VRU;
• generating a set of predictive inputs that indicate, for each of at least a subset of the plurality of environmental features, a physical relationship between the VRU and the environmental feature, wherein the physical relationship is determined with respect to the nominal heading of the VRU and the current location of the VRU;
• processing the set of predictive inputs with a heading estimation model to determine a predicted heading offset of the VRU, wherein the predicted heading offset indicates a difference between (i) a direction of a vector defined by the nominal heading of the VRU and (ii) a direction of a vector from the current location of the VRU to a predicted terminal location of the VRU's movements;
• and
• directing the vehicle to implement an autonomous driving decision generated using the predicted heading offset of the VRU.