Vehicle control system and method of use

Abstract

A system for controlling a vehicle navigating a roadway, including a perception module that generates sensor data and outputs a cost map and traffic data associated with traffic objects, a behavior planning module that receives the cost map and the traffic data from the perception module and generates planner primitives, a training module that receives the cost map and the traffic data from the perception module, receives driver input from a vehicle operator, and trains the behavior planning module, a local planning module comprising a set of task blocks that receives the cost map from the perception module and the planner primitives from the behavior planning module, selects a task block, and generates control commands using the selected task block; and a control module comprising an actuation subsystem, wherein the control module receives the control commands from the local planning module and controls the actuation subsystem.

What is claimed is: 
     
         1 . A system for controlling a vehicle navigating a roadway, comprising:
 a perception module that generates sensor data indicative of traffic objects proximal the vehicle, wherein the perception module analyzes the traffic objects to generate a cost map of the roadway;   a behavior planning module that generates planner primitives based on the cost map, wherein the behavior planning module comprises a decision-making block that comprises a probabilistic model;   a local planning module comprising a set of task blocks, each of the set of task blocks comprising a deterministic set of rules, wherein the local planning module selects a task block based on the planner primitives, and uses the deterministic set of rules of the selected task block to generate control commands based on the cost map; and   a control module comprising an actuation subsystem, wherein the control module receives the control commands from the local planning module and controls the actuation subsystem based on the control commands.   
     
     
         2 . The system of  claim 1 , wherein the sensor data comprises at least one of a set of positions and a set of trajectories, the set of positions and the set of trajectories associated with the traffic objects, wherein the traffic objects are classified as at least one of a neighboring vehicle and a static traffic object by an object analysis block of the perception module that processes the sensor data. 
     
     
         3 . The system of  claim 1 , wherein the perception module outputs a geographic location of the vehicle, and further comprising a mission planning module that receives the geographic location from the perception module and generates a route plan based on the geographic location and a destination, wherein the behavior planning module receives the route plan from the mission planning module and generates the planner primitives based on the route plan. 
     
     
         4 . The system of  claim 1 , wherein the cost map comprises a two-dimensional mapping of a set of weights to an associated set of positions on the roadway. 
     
     
         5 . The system of  claim 4 , wherein each weight of the set of weights corresponds to a risk value of occurrence of an adverse event were the vehicle to be located at the associated position. 
     
     
         6 . The system of  claim 1 , further comprising a prediction block that estimates future positions of traffic objects relative to the vehicle, based on computed trajectories of the traffic objects, wherein the behavior module generates planner primitives based on the future positions of traffic objects. 
     
     
         7 . The system of  claim 1 , further comprising a finite-state automator that selects a subset of allowed planner primitives from a set of planner primitives, based on the cost map. 
     
     
         8 . The system of  claim 7 , wherein the subset of allowed planner primitives excludes a lane-change primitive, based on sensor data indicative that traffic objects are positioned proximal a side of the vehicle. 
     
     
         9 . The system of  claim 1 , further comprising a training module that receives the cost map and the sensor data from the perception module, receives driver input from a vehicle operator, and generates the probabilistic model of the behavior planning module based on the driver input, the cost map, and the sensor data. 
     
     
         10 . The system of  claim 9 , wherein the vehicle operator is a remote vehicle operator residing outside the vehicle, further comprising a remote teleoperation interface, comprising a display and a set of inputs, that renders the sensor data to the remote vehicle operator at the display and receives the driver input from the remote vehicle operator at the set of inputs, wherein the set of inputs comprises a steering wheel input, a gas pedal input, a brake pedal input, and a transmission input. 
     
     
         11 . The system of  claim 1 , wherein each of the set of task blocks consists essentially of the deterministic set of rules. 
     
     
         12 . The system of  claim 11 , wherein the deterministic set of rules comprises an explicitly-programmed set of rules. 
     
     
         13 . The system of  claim 1 , wherein the probabilistic model comprises a trained machine-learning model. 
     
     
         14 . The system of  claim 13 , wherein the decision-making block consists essentially of the trained machine-learning model. 
     
     
         15 . A method for controlling a vehicle, comprising:
 continuously sampling, at a sensor subsystem of the vehicle, sensor data comprising an image stream, a localization signal, and operational data;   transmitting the image stream, the localization signal, and the operational data to a remote teleoperation interface associated with a teleoperator;   receiving, at a behavior planning module of the vehicle, a first directive from the teleoperator by way of the remote teleoperation interface, wherein the behavior planning module comprises a probabilistic model;   generating a planner primitive at the behavior planning module based on the directive using the probabilistic model;   selecting, at a local planning module of the vehicle, a task block based on the planner primitive, wherein the task block comprises a deterministic set of rules;   controlling the vehicle, at a control module of the vehicle, by executing the deterministic set of rules of the selected task block based on the sensor data;   receiving a second directive from the teleoperator, in response to the vehicle entering a geographic region having a predetermined characteristic;   transferring planning authority to the behavior planning module of the vehicle, in response to receiving the second directive;   automatically generating a second planner primitive at the behavior planning module based on the sensor data;   automatically selecting a second task block, wherein the second task block comprises a second deterministic set of rules, based on the second planner primitive;   controlling the vehicle by executing the second deterministic set of rules of the second task block based on the sensor data; and   automatically transferring planning authority to the teleoperator, in response to the vehicle reaching a predetermined geographic location.   
     
     
         16 . The method of  claim 15 , further comprising rendering the operational data at the remote teleoperation interface. 
     
     
         17 . The method of  claim 15 , wherein the task block consists essentially of the deterministic set of rules. 
     
     
         18 . The method of  claim 15 , wherein the probabilistic model comprises a trained machine-learning model. 
     
     
         19 . The method of  claim 18 , further comprising training the trained machine-learning model based on the first directive in combination with the sensor data. 
     
     
         20 . The method of  claim 18 , wherein the behavior planning module consists essentially of the trained machine-learning model.