EP-4735316-A1 - SYSTEMS AND METHODS FOR DECISION MAKING FOR AUTONOMOUS VEHICLES

EP4735316A1EP 4735316 A1EP4735316 A1EP 4735316A1EP-4735316-A1

Abstract

A control system for controlling a motion of an ego-vehicle traveling to a target destination is provided. The control system includes a memory and a processor to execute instruction stored by the memory. The memory stores multiple trajectory-generating functions corresponding to a maneuver defined by a parameter vector associated with a driving decision. The parameter vector is defined by one or multiple parameters. Each of the multiple trajectory-generating functions is configured to generate an achieving sequence of regions of states and values of the parameter vector reaching an input target region within a prediction horizon. The stored instructions cause the control system to test control admissibility of at least some of the driving decisions consistent with the target destination of the ego-vehicle at a current state. The stored instruction also caused the control system to control the ego-vehicle according to one of the admissible driving decisions.

Inventors

DI CAIRANO, STEFANO
Vinod, Abraham
SKIBIK, Terrence
WEISS, AVISHAI
BERNTORP, KARL

Assignees

Mitsubishi Electric Mobility Corporation

Dates

Publication Date: 20260506
Application Date: 20240509

Claims (20)

[Claim 1] A control system configured for controlling a motion of an ego-vehicle traveling to a target destination, comprising: a memory configured to store multiple trajectory-generating functions, wherein each of the multiple trajectory-generating functions corresponds to a maneuver defined by a parameter vector associated with a driving decision of a plurality of driving decisions, wherein the parameter vector is defined by one or multiple parameters, wherein each of the multiple trajectory-generating functions is configured to generate a sequence of states within a prediction horizon, from a given value of the parameter vector and a current state of the ego- vehicle, wherein at least two of the multiple trajectory-generating functions are configured for different types of parameters of the parameter vector; and at least one processor coupled with stored instructions, which when executed by the processor, cause the control system to test control admissibility of at least some of the driving decisions consistent with the target destination of the ego-vehicle at the current state, wherein, to test the driving decision, the processor is configured to: map the tested driving decision to a target region of states of the ego-vehicle at an instance of time within the prediction horizon; collect one or multiple obstacle states of one or multiple obstacles proximate to the ego-vehicle at one or multiple instances of time within the prediction horizon; construct the sequence of achieving regions of states and values of the parameter vector reaching the target region of states of the egovehicle within the prediction horizon for the trajectory-generating function associated with the tested driving decision for the target region of states of the ego-vehicle; construct one or multiple sequences of colliding regions of states and values of the parameter vector reaching each of the one or multiple obstacle states within the prediction horizon for the trajectory-generating function associated with the tested driving decision for each of the one or more obstacle states; declare the driving decision as the admissible driving decision if there is at least one of the one or multiple values of the parameter vector for which the current state of the ego-vehicle is inside the achieving sequence of regions of states reaching the target region of states of the ego-vehicle and outside of all of the sequences of regions of states reaching each one or multiple obstacle states; and control the ego-vehicle according to a motion model corresponding to the admissible driving decisions to reach the target region of states of at least one admissible driving decision.
[Claim 2] The control system of claim 1, wherein at least one value of the one or multiple parameters of the parameter vector is a steady state target value for the motion generated by the trajectory-generating function, and wherein the one or multiple parameters of the parameter vector comprise target lateral displacement, target longitudinal displacement, target velocity, and/or target heading.
[Claim 3] The control system of claim 1, wherein at least one of the trajectorygenerating functions are generated by fitting motion data according to a parametric dynamical model.
[Claim 4] The control system of claim 1, wherein at least one of the multiple trajectory-generating functions is obtained by combining a motion model with a vehicle controller model, and wherein a value of the parameter vector associated with the at least one of the multiple trajectory-generating functions is a setpoint of the control system.
[Claim 5] The control system of claim 1, wherein the driving decision comprises changing lanes, following a lane, stopping, turning left, or turning right.
[Claim 6] The control system of claim 1, wherein the current state of the egovehicle comprises a current location and at least one of a current velocity, a current acceleration, and a current heading.
[Claim 7] The control system of claim 6, wherein the target region of states comprises a target region of positions and at least some of a target velocity, a target acceleration, and/or a target heading.
[Claim 8] The control system of claim 7, wherein the target region is represented as a target polyhedron about a target center location, and wherein the target center location moves according to a motion prediction model, and wherein the achieving sequence of achieving regions of states and values of the parameter vector is constructed such that if a parameter value of the achieving sequence is applied to a corresponding state value, the target region is entered within a fixed number of achieving steps, and wherein the fixed number of achieving steps is shorter than a length of the maneuver which is shorter or equal to the prediction horizon.
[Claim 9] The control system of claim 1, wherein at least one of the one or multiple obstacle states comprise an obstacle region, and at least some of an obstacle velocity value, an obstacle acceleration value, and/or an obstacle yaw rate value.
[Claim 10] The control system of claim 9, wherein the obstacle region is represented as an obstacle polyhedron about an obstacle center location, and wherein the obstacle center location moves according to a motion prediction model, and wherein at least one of the colliding sequences of colliding regions of states and values of the parameter vector is constructed such that if a parameter value of at least one of the colliding sequences is applied to a corresponding state value, the obstacle region is entered within a fixed number of colliding steps, and wherein the fixed number of colliding steps is shorter than a length of the maneuver.
[Claim 11] The control system of claim 1, wherein at least one of the sequence of achieving regions of states and/or the one or multiple sequences of colliding regions of states associated to at least one of the multiple trajectory-generating functions are computed prior to vehicle system operation and stored in the memory.
[Claim 12] The control system of claim 1, wherein, upon determining one value of the parameter vector for one driving decision determined to be an admissible driving decision, the control system immediately stops testing others of the at least some of the driving decisions.
[Claim 13] The control system of claim 1, wherein the processor is further configured to determine all values of the parameter vector with a given value range that corresponding to admissible driving decisions.
[Claim 14] The control system of claim 1 , wherein the driving decision is declared as the admissible driving decision by evaluating upper bound inequalities determining one or more regions within the achieving regions and outside of the colliding regions, wherein the upper bound inequalities are evaluated by: inserting values corresponding to the current state of the ego-vehicle into the upper bound inequalities to obtain reduced order inequalities; removing redundant inequalities; and determining one or more values of the parameter vector that satisfy the reduced order inequalities.
[Claim 15] The control system of 14, wherein the reduced order inequalities comprise a first inequality and a second inequality, wherein the first inequality is obtained by retaining one of the reduced order inequalities with the largest ratio of constant term and positive coefficient of the value of the current state, and wherein the second inequality is obtained by retaining one of the reduced order inequalities with the smallest ratio of constant term and positive coefficient of the value of the current state.
[Claim 16] The control system of claim 1, wherein the admissible driving decision is selected according to a maneuver priority level, wherein the maneuver priority level corresponds to a desirability of the maneuver for driver comfort and driver performance.
[Claim 17] The control system of claim 1 , wherein a value of the parameter vector is determined by optimizing a cost function subject to constraints of the value of the parameter vector that must result in an admissible driving decision, wherein the cost function includes at least one of the following terms: maneuver velocity, maneuver completion time, driver comfort, maneuver aggressiveness, parameter value amplitude, and maneuver robustness, and wherein the maneuver robustness is defined as the maximum value that any perturbation of amplitude equal or smaller than such maximum value still results in an admissible maneuver.
[Claim 18] The control system of claim 1, wherein the processor controls egovehicle by providing a motion planner of the ego-vehicle the maneuver and the target region of states corresponding to the admissible driving decision, wherein the processor further provides the motion planner with a set of achieving regions providing a suggested trajectory to achieve the target region of states, wherein the suggested trajectory is computed by the trajectorygenerating function of the maneuver corresponding to the admissible driving decision for the value of the parameter vector achieving the target region of states from the current state of the vehicle.
[Claim 19] A method for controlling a motion of an ego-vehicle traveling to a target destination, comprising: mapping, via at least one processor, a driving decision to a target region of states of the ego-vehicle at an instance of time within a prediction horizon; collecting, via the at least one processor, one or multiple obstacle states of one or multiple obstacles proximate to the ego-vehicle at one or multiple instances of time within the prediction horizon; constructing, via the at least one processor, a sequence of achieving regions of states and values of a parameter vector reaching the target region of states of the ego-vehicle within the prediction horizon for one of multiple trajectory-generating functions associated with the driving decision for the target region of states of the ego-vehicle, wherein each of the multiple trajectory-generating functions corresponds to a maneuver defined by the parameter vector associated with the driving decision; constructing, via the at least one processor, one or multiple sequences of colliding regions of states and values of the parameter vector reaching each of the one or multiple obstacle states within the prediction horizon for the trajectory-generating function associated within the driving decision for each of the one or multiple obstacle states; declaring, via the at least one processor, the driving decision as an admissible driving decision if there is at least one of the one or multiple values of the parameter vector for which a current state of the ego-vehicle is inside the achieving sequence of regions of states reaching the target region of states of the ego-vehicle and outside of all of the colliding sequences of regions of states reaching each one or multiple obstacle states; and controlling, via the at least one processor, the ego-vehicle according to the admissible driving decisions to reach the target region of states of at least one admissible driving decision.
[Claim 20] A non-transitory computer readable memory embodied thereon a program executable by at least one processor for performing a method for controlling a motion of an ego-vehicle traveling to a target destination, comprising: mapping, via at least one processor, a driving decision to a target region of states of the ego-vehicle at an instance of time within a prediction horizon; collecting, via the at least one processor, one or multiple obstacle states of one or multiple obstacles proximate to the ego-vehicle at one or multiple instances of time within the prediction horizon; constructing, via the at least one processor, a sequence of achieving regions of states and values of a parameter vector reaching the target region of states of the ego-vehicle within the prediction horizon for one of multiple trajectory-generating functions associated with the driving decision for the target region of states of the ego-vehicle, wherein each of the multiple trajectory-generating functions corresponds to a maneuver defined by the parameter vector associated with the driving decision; constructing, via the at least one processor, one or multiple sequences of colliding regions of states and values of the parameter vector reaching each of the one or multiple obstacle states within the prediction horizon for the trajectory-generating function associated within the driving decision for each of the one or multiple obstacle states; declaring, via the at least one processor, the driving decision as an admissible driving decision if there is at least one of the one or multiple values of the parameter vector for which a current state of the ego-vehicle is inside the achieving sequence of regions of states reaching the target region of states of the ego-vehicle and outside of all of the colliding sequences of regions of states reaching each one or multiple obstacle states; and controlling, via the at least one processor, the ego-vehicle according to the admissible driving decisions to reach the target region of states of at least one admissible driving decision.

Description

[DESCRIPTION] [Title of Invention] SYSTEMS AND METHODS FOR DECISION MAKING FOR AUTONOMOUS VEHICLES [Technical Field] [0001] The present disclosure relates generally to autonomous driving and advanced driver-assistance systems, and, more particularly, to evaluating driving decisions of an ego-vehicle taking into account the current state of the vehicle and the behavior of nearby obstacles. [Background Art] [0002] Conventional autonomously driving vehicles must be equipped with a control system that determines how the vehicle should move on the road, accounting for motor vehicle laws and traffic, to achieve its driving objectives. Conventional control systems determine vehicle motion by analyzing the environment based on data acquired by a plurality of sensors and processed by recognition and mapping algorithms, by computing a desired vehicle path and speed based on this environmental data, and by controlling the vehicle to follow that path using available vehicle actuators. Due to the complexity of such conventional operations, the conventional control systems include both path planning and vehicle control subsystems. For instance, US Patent No. 9,915,948 discusses how the vehicle control and the path planning subsystems can be integrated to guarantee that the vehicle achieves a desired driving objective. [0003] Some methods have been proposed in the prior art for determining goals (or targets) of a path planning system of an autonomous vehicle that ensure that each intermediate goal can be achieved by a vehicle motion planned by a path planner in accordance with the traffic and driving rules, and that achieving a sequence of intermediate goals, leads to achieving the overall driving objectives. For instance, US Patent No. 10,860,023 proposes determining the intermediate goal that can be achieved by evaluating whether the state of the vehicle is contained in appropriate regions of space from which there exists trajectories that reach the goals, and then evaluate whether a candidate trajectory can be generated that does not collide with obstacles, including other vehicles, that are present in the road. Thus, while the method in US Patent 10,860,023 is appealing, it suffers from the limitation of requiring generating candidate trajectories online to check achievability of the goal while avoiding obstacles, and such check also requires collision-checking on a number of points along the trajectory. Both of these operations are computationally expensive as several trajectories need to be generated and multiple points of such trajectories need to be evaluated to determine if these points will collide with obstacles present on the road. [0004] Thus, there is a need for systems and methods for determining goals for a path planning system of an autonomous vehicle that reduce the computations for ensuring that each intermediate target can be achieved by a vehicle motion planned by a path planner in accordance with the traffic and driving rules without colliding with obstacles. [Summary of Invention] [0005] Some embodiments of the present disclosure are based on recognizing that there is a finite set of decisions an ego-vehicle can make while moving toward a final destination. For instance, in city driving, this set of decisions includes most driving tasks, such as whether to stay in a lane or change lanes, whether to maintain a speed or accelerate/decelerate, and/or whether to yield to other vehicles at intersections. Thus, at least one realization is that while there are a finite set of decisions a vehicle can make, there is also an infinite number of trajectories along which the vehicle may move and still satisfy the same decision. Determining all of the trajectories a vehicle can follow to result in a certain decision, such as a lane change, while also determining whether the trajectories may be followed without colliding with another vehicle, is a daunting task and computationally expensive. [0006] The term “ego-vehicle” as used herein refers to a self-driving (i.e. autonomous) or semi-autonomous vehicle equipped with sensor systems, computing devices, and control mechanisms that enable it to perceive its environment, process data, and make decisions without direct human intervention. The ego-vehicle is capable of autonomously navigating through a road or traffic environment, detecting and avoiding obstacles, and following traffic regulations, thereby providing transportation services with limited or no human involvement. [0007] Some embodiments of the present disclosure are based on the realization that determining whether a certain decision is to be made is dependent on whether that decision can be accomplished by a safe trajectory of the vehicle. However, determining such trajectories explicitly is difficult to implement in real time due to the computational limitations. Consequently, several embodiments of the present disclosure are based on the understanding that determining whether certain decisions lead