Hybrid data-driven actor prediction using reachable sets and supervised learning
A method for predicting an actor occupancy corridor includes receiving input data, predicting an occupancy sets of the actor using reachability analysis and the input data, determining an occupancy corridor constraints of the actor using a machine learning model and the input data, and determining an occupancy corridor of the actor using the occupancy corridor constraints of the actor using the machine learning model and the occupancy sets of the actor using a reachability analysis. Moreover, the method includes controlling the movement of a host vehicle based on the occupancy corridor of the actor.
1 . A method for predicting an actor occupancy corridor, comprising:
receiving input data, wherein the input data includes map data and initial trajectory data, the map data includes a map of an area surrounding an actor, and the initial trajectory data includes an initial trajectory of the actor;
predicting a plurality of occupancy sets of the actor over a predetermined period of time in a future using a reachability theory analysis and the input data;
determining occupancy corridor constraints of the actor using a machine learning model and the input data, wherein a prediction horizon length of the occupancy set prediction is a function of the estimated time to collision to the nearest actor and a nominal prediction hypothesis uncertainty;
providing the occupancy sets and the nominal machine learning predictions to a predictor resolver, wherein the predictor resolver blends the nominal machine learning predictions and the occupancy sets together and corrects the nominal predictions that are infeasible via projection and convex optimization, thereby further determining the occupancy corridor constraints;
determining an occupancy corridor of the actor using the occupancy corridor constraints of the actor using the machine learning model and the occupancy sets of the actor using a reachability analysis; and
controlling a movement of a host vehicle based on the occupancy corridor of the actor.
2 . The method of claim 1 , further comprising:
receiving perception data from a plurality of sensors of the host vehicle, wherein the perception data includes information about the area surrounding the actor; and
fusing the perception data with the map data to generate scene data, wherein the scene data includes a nominal route path and a road centerline of a road, and the actor is moving on the road.
3 . The method of claim 2 , wherein the initial trajectory data includes a longitudinal acceleration and a lateral acceleration of the actor, the reachability analysis uses the longitudinal acceleration and the lateral acceleration of the actor, and the occupancy sets of the actor is a forward reachable set over the predetermined period of time.
4 . The method of claim 3 , wherein predicting the occupancy sets of the actor further comprises:
determining that the forward reachable set of the actor does not intersect other traffic participants for the predetermined period of time; and
determining that the forward reachable set of the actor is within permissible road positions.
5 . The method of claim 4 , further comprising:
using a reasoner to determine output accelerations of the actor using the input data;
wherein the occupancy corridor is determined using the output accelerations of the actor.
6 . The method of claim 5 , wherein the occupancy corridor constraints of the actor include maximum acceleration of the actor and minimum acceleration of the actor.
7 . The method of claim 6 , further comprising determining that a trajectory of the host vehicle intersects the occupancy corridor of the actor.
8 . The method of claim 7 , further comprising removing the trajectory of the host vehicle that intersects the occupancy corridor of the actor.
9 . The method of claim 6 , further comprising determining that a trajectory of the host vehicle does not intersect the occupancy corridor of the actor.
10 . The method of claim 9 , wherein controlling the movement of the host vehicle based on the occupancy corridor of the actor includes moving along the trajectory of the host vehicle that does not intersect the occupancy corridor of the actor.
11 . A system for predicting an actor occupancy corridor, comprising:
a plurality of sensors;
a controller in communication with the plurality of sensors, wherein the controller includes a processor and a non-transitory computer-readable medium, and the controller is programmed to:
receive input data, wherein the input data includes map data and initial trajectory data, the map data includes a map of an area surrounding an actor, and the initial trajectory data includes an initial trajectory of the actor;
predict a plurality of occupancy sets of the actor over a predetermined period of time using reachability analysis and the input data;
determine occupancy corridor constraints of the actor using a machine learning model and the input data, wherein a prediction horizon length of the occupancy set prediction is a function of the estimated time to collision to the nearest actor and a nominal prediction hypothesis uncertainty;
provide the occupancy sets and the nominal machine learning predictions to a predictor resolver, wherein the predictor resolver blends the nominal machine learning predictions and the occupancy sets together and corrects the nominal predictions that are infeasible via projection and convex optimization, thereby further determining the occupancy corridor constraints;
determine an occupancy corridor of the actor using the occupancy corridor constraints of the actor using a machine learning model and the plurality of occupancy sets of the actor using reachability analysis; and
control a movement of a host vehicle based on the occupancy corridor of the actor.
12 . The system of claim 11 , wherein the controller is programmed to:
receive perception data from a plurality of sensors of the host vehicle, wherein the perception data includes information about the area surrounding the actor; and
fuse the perception data with the map data to generate scene data, wherein the scene data includes a nominal route path and a road centerline of a road, and the actor is moving on the road.
13 . The system of claim 12 , wherein the initial trajectory data includes a longitudinal acceleration and a lateral acceleration of the actor, the reachability analysis uses the longitudinal acceleration and the lateral acceleration of the actor, and the occupancy sets of the actor is a forward reachable set over the predetermined period of time.
14 . The system of claim 13 , wherein the controller is programmed to:
determine that the forward reachable set of the actor does not intersect other traffic participants for the predetermined period of time; and
determine that the forward reachable set of the actor is within permissible road positions.
15 . The system of claim 14 , wherein the controller is programmed to:
use a reasoner to determine output accelerations of the actor using the input data;
wherein the occupancy corridor is determined using the output accelerations of the actor.
16 . The system of claim 15 , wherein the occupancy corridor constraints of the actor include maximum acceleration of the actor and minimum acceleration of the actor.
17 . The system of claim 16 , wherein the controller is programmed to determine that a trajectory of the host vehicle intersects the occupancy corridor of the actor.
18 . The system of claim 17 , wherein the controller is programmed to remove the trajectory of the host vehicle that intersects the occupancy corridor of the actor.
19 . The system of claim 16 , wherein the controller is programmed to determine that a trajectory of the host vehicle does not intersect the occupancy corridor of the actor.
20 . A tangible, non-transitory, machine-readable medium, comprising machine-readable instructions, that when executed by a processor, cause the processor to:
receive input data, wherein the input data includes map data and initial trajectory data, the map data includes a map of an area surrounding an actor, and the initial trajectory data includes an initial trajectory of the actor;
predict a plurality of occupancy sets of the actor over a predetermined period of time in a future using reachability analysis and the input data;
determine occupancy corridor constraints of the actor using a machine learning model and the input data, wherein a prediction horizon length of the occupancy set prediction is a function of the estimated time to collision to the nearest actor and a nominal prediction hypothesis uncertainty;
provide the occupancy sets and the nominal machine learning predictions to a predictor resolver, wherein the predictor resolver blends the nominal machine learning predictions and the occupancy sets together and corrects the nominal predictions that are infeasible via projection and convex optimization, thereby further determining the occupancy corridor constraints;
determine an occupancy corridor of the actor using the occupancy corridor constraints of the actor using a machine learning model and the occupancy sets of the actor using reachability analysis; and
control a movement of a host vehicle based on the occupancy corridor of the actor.