Combined prediction and path planning for autonomous objects using neural networks

Abstract

Sensors measure information about actors or other objects near an object, such as a vehicle or robot, to be maneuvered. Sensor data is used to determine a sequence of possible actions for the maneuverable object to achieve a determined goal. For each possible action to be considered, one or more probable reactions of the nearby actors or objects are determined. This can take the form of a decision tree in some embodiments, with alternative levels of nodes corresponding to possible actions of the present object and probable reactive actions of one or more other vehicles or actors. Machine learning can be used to determine the probabilities, as well as to project out the options along the paths of the decision tree including the sequences. A value function is used to generate a value for each considered sequence, or path, and a path having a highest value is selected for use in determining how to navigate the object.

What is claimed is: 
     
         1 . A computer-implemented method, comprising:
 sensing, using one or more sensors of a first object, one or more characteristics of one or more secondary objects;   determining, using a processor of the first object, and based on the one or more characteristics and probable reactive actions of at least one second object, one or more possible navigation paths for the one or more secondary objects; and   selecting a navigation path from the one or more possible navigation paths based, at least in part, on a value function corresponding to sensed characteristics of the one or more secondary objects.   
     
     
         2 . The computer-implemented method of  claim 1 , wherein the one or more characteristics include at least one of a position, a velocity, a rate of acceleration, a direction of motion, or a motion characterization. 
     
     
         3 . The computer-implemented method of  claim 1 , further comprising:
 providing at least a first action of the selected navigation path to an optimizer of the first object; and   causing the first object to maneuver based upon navigation instructions generated by the optimizer.   
     
     
         4 . The computer-implemented method of  claim 1 , further comprising:
 generating a decision tree for the first object, the decision tree including alternating levels of possible actions for the first object and the probable reactive actions of the at least one secondary object.   
     
     
         5 . The computer-implemented method of  claim 4 , further comprising:
 utilizing a policy network to determine probable responsive actions at levels of the decision tree for consideration for the one or more possible navigation paths.   
     
     
         6 . The computer-implemented method of  claim 4 , further comprising:
 using a trained neural network to infer the probable reactive actions at levels of the decision tree.   
     
     
         7 . The computer-implemented method of  claim 1 , further comprising:
 using at least one move generator to determine the probable reactive actions of the at least one secondary object, the at least one move generator including a trained neural network for a characterization of the at least one secondary object.   
     
     
         8 . The computer-implemented method of  claim 1 , further comprising:
 determining characterizations for the at least one secondary object, the characterizations determining probabilities for the probable reactive actions of at least one secondary object.   
     
     
         9 . The computer-implemented method of  claim 1 , further comprising:
 determining criticality values for the at least one secondary object, a number of probable reactive actions for the secondary objects determined based at least in part upon the criticality values.   
     
     
         10 . A system, comprising:
 one or more processors; and   memory including instructions that, if performed by the one or more processors, cause the system to:
 sense, using one or more sensors of a first object, one or more characteristics of one or more secondary objects; 
 determine, using a processor of the first object, and based on the one or more characteristics and probable reactive actions of at least one second object, one or more possible navigation paths for the one or more secondary objects; and 
 select a navigation path from the one or more possible navigation paths based, at least in part, on a value function corresponding to sensed characteristics of the one or more secondary objects. 
   
     
     
         11 . The system of  claim 10 , wherein the one or more characteristics include at least one of a position, a velocity, a rate of acceleration, a direction of motion, or a motion characterization. 
     
     
         12 . The system of  claim 10 , wherein the instructions if performed further cause the system to:
 provide at least a first action of the selected navigation path to an optimizer of the first object; and   cause the first object to maneuver based upon navigation instructions generated by the optimizer.   
     
     
         13 . The system of  claim 10 , wherein the instructions if performed further cause the system to:
 generate a decision tree for the first object, the decision tree including alternating levels of possible actions for the first object and the probable reactive actions of the at least one secondary object.   
     
     
         14 . The system of  claim 13 , wherein the instructions if performed further cause the system to:
 utilize a policy network to determine probable responsive actions at levels of the decision tree for consideration for the one or more possible navigation paths.   
     
     
         15 . The system of  claim 13 , wherein the instructions if performed further cause the system to:
 use a trained neural network to infer the probable reactive actions at levels of the decision tree.   
     
     
         16 . The system of  claim 10 , wherein the instructions if performed further cause the system to:
 use at least one move generator to determine the probable reactive actions of the at least one secondary object, the at least one move generator including a trained neural network for a characterization of the at least one secondary object.   
     
     
         17 . The system of  claim 10 , wherein the instructions if performed further cause the system to:
 determine characterizations for the at least one secondary object, the characterizations determining probabilities for the probable reactive actions of at least one secondary object.   
     
     
         18 . The system of  claim 10 , wherein the instructions if performed further cause the system to:
 determine criticality values for the at least one secondary object, a number of probable reactive actions for the secondary objects determined based at least in part upon the criticality values.   
     
     
         19 . A navigation system, comprising:
 one or more sensors to sense one or more characteristics of one or more objects;   one or more processors to:
 determine, based on the one or more characteristics and probable reactive actions of at least one object, one or more possible navigation paths for the one or more objects; and 
 select a navigation path from the one or more possible navigation paths based, at least in part, on a value function corresponding to sensed characteristics of the one or more objects. 
   
     
     
         20 . The navigation system of  claim 19 , wherein the one or more processors are further to:
 provide at least a first action of the selected navigation path to an optimizer; and   cause a maneuver to be performed based at least in part upon navigation instructions generated by the optimizer.