Predicting Jaywaking Behaviors of Vulnerable Road Users

Abstract

Jaywalking behaviors of vulnerable road users (VRUs) such as cyclists or pedestrians can be predicted. Location data is obtained that identifies a location of a VRU within a vicinity of a vehicle. Environmental data is obtained that describes an environment of the VRU, where the environmental data identifies a set of environmental features in the environment of the VRU. The system can determine a nominal heading of the VRU, and generate a set of predictive inputs that indicate, for each of at least a subset of the set of environmental features, a physical relationship between the VRU and the environmental feature. The physical relationship can be determined with respect to the nominal heading of the VRU and the location of the VRU. The set of predictive inputs can be processed with a heading estimation model to generate a predicted heading offset (e.g., a target heading offset) for the VRU.

1 - 20 . (canceled) 
     
     
         21 . A method performed by a system of one or more computers, comprising:
 identifying a current location of an agent detected in a vicinity of a vehicle traveling on a roadway;   determining a nominal heading of the agent;   obtaining a plurality of environmental features that describe information about an environment of the agent;   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 agent and the environmental feature, wherein the physical relationship is determined with respect to the nominal heading of the agent and the current location of the agent;   processing the set of predictive inputs with a heading estimation model to generate a predicted heading of the agent, wherein the predicted heading is generated by determining a direction from the current location of the agent to a predicted terminal location of a movement of the agent; and   using the predicted heading of the agent to control a movement of the vehicle.   
     
     
         22 . The method of  claim 21 , comprising assigning a current heading of the agent as the nominal heading of the agent. 
     
     
         23 . The method of  claim 21 , comprising:
 selecting a first environmental feature of the plurality of environmental features; and   determining a location of a point along the first environmental feature;   wherein determining the nominal heading of the agent comprises determining a vector between the agent and the point along the first environmental feature, and assigning a direction of the vector as the nominal heading of the agent.   
     
     
         24 . The method of  claim 23 , wherein:
 the first environmental feature is a road edge;   the point along the first environmental feature is selected based on being the closest point along the road edge to the agent; and   the nominal heading indicates a direction from the agent to the closest point along the road edge.   
     
     
         25 . The method of  claim 21 , comprising:
 determining, based on a location history of the agent and environmental data, a jaywalking prediction that indicates whether the agent is jaywalking or is likely to jaywalk on the roadway;   wherein the system generates the predicted heading for the agent responsive to the jaywalking prediction indicating that the agent is jaywalking or is likely to jaywalk on the roadway.   
     
     
         26 . The method of  claim 25 , wherein the system is configured not to generate a predicted heading for the agent responsive to the jaywalking prediction indicating that the agent is not jaywalking and is not likely to jaywalk on the roadway traveled by the vehicle. 
     
     
         27 . The method of  claim 21 , wherein the jaywalking prediction model comprises a decision tree, a random decision forest, an artificial neural network, or a regression model. 
     
     
         28 . The method of  claim 21 , wherein the plurality of environmental features include at least one of a road edge, a lane boundary, a sidewalk, a bicycle lane, a road curb, or an intersection. 
     
     
         29 . The method of  claim 21 , wherein the agent is a pedestrian, a cyclist, or a low-speed motorized vehicle, wherein the vehicle is a fully autonomous or semi-autonomous vehicle. 
     
     
         30 . The method of  claim 21 , wherein the physical relationship between the agent and a first environmental feature of the plurality of environmental features, as indicated by a first predictive input of the set of predictive inputs, comprises at least one of a positional relationship, a distal relationship, or an angular relationship between the agent and the first environmental feature. 
     
     
         31 . A system, comprising:
 a data processing apparatus; and   one or more non-transitory computer-readable media encoded with instructions that, when executed by the data processing apparatus, cause performance of operations comprising:
 identifying a current location of an agent detected in a vicinity of a vehicle traveling on a roadway; 
 determining a nominal heading of the agent; 
 obtaining a plurality of environmental features that describe information about an environment of the agent; 
 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 agent and the environmental feature, wherein the physical relationship is determined with respect to the nominal heading of the agent and the current location of the agent; 
 processing the set of predictive inputs with a heading estimation model to generate a predicted heading of the agent, wherein the predicted heading is generated by determining a direction from the current location of the agent to a predicted terminal location of a movement of the agent; and 
 using the predicted heading of the agent to control a movement of the vehicle. 
   
     
     
         32 . The system of  claim 31 , wherein the operations comprise assigning a current heading of the agent as the nominal heading of the agent. 
     
     
         33 . The system of  claim 31 , wherein the operations comprise:
 selecting a first environmental feature of the plurality of environmental features; and   determining a location of a point along the first environmental feature;   wherein determining the nominal heading of the agent comprises determining a vector between the agent and the point along the first environmental feature, and assigning a direction of the vector as the nominal heading of the agent.   
     
     
         34 . The system of  claim 33 , wherein:
 the first environmental feature is a road edge;   the point along the first environmental feature is selected based on being the closest point along the road edge to the agent; and   the nominal heading indicates a direction from the agent to the closest point along the road edge.   
     
     
         35 . The system of  claim 31 , wherein the operations comprise:
 determining, based on a location history of the agent and environmental data, a jaywalking prediction that indicates whether the agent is jaywalking or is likely to jaywalk on the roadway;   wherein the system generates the predicted heading for the agent responsive to the jaywalking prediction indicating that the agent is jaywalking or is likely to jaywalk on the roadway.   
     
     
         36 . The system of  claim 35 , wherein the operations comprise selecting not to generate a predicted heading for the agent responsive to the jaywalking prediction indicating that the agent is not jaywalking and is not likely to jaywalk on the roadway traveled by the vehicle. 
     
     
         37 . The system of  claim 31 , wherein the jaywalking prediction model comprises a decision tree, a random decision forest, an artificial neural network, or a regression model. 
     
     
         38 . The system of  claim 31 , wherein the plurality of environmental features include at least one of a road edge, a lane boundary, a sidewalk, a bicycle lane, a road curb, or an intersection. 
     
     
         39 . The system of  claim 31 , wherein the agent is a pedestrian, a cyclist, or a low-speed motorized vehicle, wherein the vehicle is a fully autonomous or semi-autonomous vehicle. 
     
     
         40 . One or more non-transitory computer-readable media encoded instructions that, when executed by data processing apparatus, cause the data processing apparatus to perform operations comprising:
 identifying a current location of an agent detected in a vicinity of a vehicle traveling on a roadway;   determining a nominal heading of the agent;   obtaining a plurality of environmental features that describe information about an environment of the agent;   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 agent and the environmental feature, wherein the physical relationship is determined with respect to the nominal heading of the agent and the current location of the agent;   processing the set of predictive inputs with a heading estimation model to generate a predicted heading of the agent, wherein the predicted heading is generated by determining a direction from the current location of the agent to a predicted terminal location of a movement of the agent; and   using the predicted heading of the agent to control a movement of the vehicle.