US 12,134,409 B1
Methods for cooperative decision-making on lane-changing behaviors of autonomous vehicles based on bayesian game
Kang Song, Tianjin (CN); Fan Guo, Tianjin (CN); and Hui Xie, Tianjin (CN)
Assigned to TIANJIN UNIVERSITY, Tianjin (CN)
Filed by TIANJIN UNIVERSITY, Tianjin (CN)
Filed on Jun. 25, 2024, as Appl. No. 18/752,907.
Application 18/752,907 is a continuation in part of application No. PCT/CN2023/086547, filed on Apr. 6, 2023.
Claims priority of application No. 202210603381.7 (CN), filed on May 30, 2022.
Int. Cl. B60W 60/00 (2020.01); B60W 50/00 (2006.01)
CPC B60W 60/0027 (2020.02) [B60W 50/0097 (2013.01); B60W 60/001 (2020.02); B60W 2050/0031 (2013.01); B60W 2050/0052 (2013.01); B60W 2552/10 (2020.02); B60W 2554/4041 (2020.02); B60W 2554/4046 (2020.02); B60W 2720/106 (2013.01)] 8 Claims
OG exemplary drawing
 
1. A method for cooperative decision-making on lane-changing behavior of an autonomous vehicle based on Bayesian game, the method comprising:
step 1, establishing a prior probability distribution of a vehicle driving style of a side vehicle, including: obtaining vehicle driving data through an intelligent networked roadside sensor, and recording and counting the prior probability distribution of the vehicle driving style under different time periods and different road sections, wherein the vehicle driving style of the side vehicle includes an aggressive driving style and a non-aggressive driving style;
step 2, outputting a lane-changing willingness through a lane-changing willingness determination module, including: collecting vehicle information of a specified vehicle and a surrounding vehicle through an on-board sensor, determining an original predetermined distance and a lane-changing predetermined distance, determining a lane-changing necessity and a lane-changing safety through a cumulative distribution function constructed by introducing an expectation and a variance, establishing a lane-changing willingness output model based on fuzzy logic, and executing following steps 3-7 in response to the lane-changing willingness reaching a lane-changing willingness threshold;
step 3, inferring a posterior probability of a vehicle driving style of a rear vehicle in a target lane using Bayesian filtering, including: in response to the lane-changing willingness of the specified vehicle reaching the lane-changing willingness threshold, determining a likelihood function of the vehicle driving style of the rear vehicle in the target lane based on an acceleration of the rear vehicle in the target lane acquired by the on-board sensor, and obtaining, through the likelihood function and the prior probability distribution, the posterior probability of the vehicle driving style of the rear vehicle in the target lane and a driver aggressiveness factor β of the rear vehicle in the target lane, the driver aggressiveness factor having a value in a range of [0, 1];
step 4, predicting, through a Long Short-Term Memory (LSTM) neural network and a vehicle kinematic model, driving trajectories, driving speeds, and accelerations of the specified vehicle and the rear vehicle in the target lane in a future-projected time domain;
step 5, establishing payoff matrices and determining a probability of lane-changing execution, including: establishing payoff matrices for a non-cooperative game, wherein the payoff matrices for the non-cooperative game include a payoff matrix of the specified vehicle and an aggressive rear vehicle in the target lane, and a payoff matrix of the specified vehicle and a non-aggressive rear vehicle in the target lane, respectively, and payoff functions of the payoff matrices include a function of a safety prediction payoff, a function of a time prediction payoff, a function of a comfort prediction payoff, and a function of a cooperation prediction payoff, and obtaining the probability of lane-changing execution by solving the payoff matrix;
step 6, updating a vehicle state, including: in response to a determination that the probability of lane-changing execution is less than an execution threshold, the specified vehicle not executing a lane-changing, updating a longitudinal trajectory of the specified vehicle; and in response to a determination that the probability of lane-changing execution is greater than or equal to the execution threshold, updating a lane-changing trajectory and the longitudinal trajectory of the specified vehicle, simultaneously;
step 7, cyclically executing dynamic game-based lane-changing decision-making, including: cyclically executing steps 3-6 until an execution of a lane-changing strategy is completed or the lane-changing willingness disappears;
wherein, in the step 5, the payoff matrix of the specified vehicle and the aggressive rear vehicle in the target lane is: (U11,Q11), (U12,Q12), (U21,Q21), and (U22,Q22), wherein
U11, U12, U21, U22 denote payoffs of the specified vehicle under a combination of four strategies with the aggressive rear vehicle in the target lane, which includes a strategy of [lane-changing, deceleration], a strategy of [lane-changing, acceleration], a strategy of [no lane-changing, deceleration], and a strategy of [no lane-changing, acceleration], respectively;
Q11,Q12,Q21,Q22 denote payoffs of the aggressive rear vehicle in the target lane under a combination of four strategies with the specified vehicle, which includes a strategy of [lane-changing, deceleration], a strategy of [lane-changing, acceleration], a strategy of [no lane-changing, deceleration], and a strategy of [no lane-changing, acceleration], respectively;
the payoff matrix of the specified vehicle and the non-aggressive rear vehicle in the target lane is: (U33,Q33), (U34,Q34), (U43,Q43), and (U44,Q44), wherein
U33, U34, U43, U44 denote payoffs of the specified vehicle under a combination of four strategies with the non-aggressive rear vehicle in the target lane, which includes a strategy of [lane-changing, deceleration], a strategy of [lane-changing, acceleration], a strategy of [no lane-changing, deceleration], and a strategy of [no lane-changing, acceleration], respectively;
Q33,Q34,Q43, Q44 denote payoffs of the non-aggressive rear vehicle in the target lane under a combination of four strategies with the specified vehicle, which includes a strategy of [lane-changing, deceleration], a strategy of [lane-changing, acceleration], a strategy of [no lane-changing, deceleration], and a strategy of [no lane-changing, acceleration], respectively;
the payoff U for the specified vehicle and the payoff Q for the rear vehicle in the target lane include payoffs at a future moment, the payoffs at the future moment including four components:
(1) the safety prediction payoff, denotes as:
Term(sf)=−{ω11[Ac(t′)+vSV(t′)*vRV(t′)]*I(Ac)+ω12[As(t′)+vSV(t′)*vRV(t′)]*I(As)}
wherein, vSV(t′) and vRV(t′) denote a driving speed of the specified vehicle and a driving speed of the rear vehicle in the target lane at a predicted moment t′, respectively, Ac(t′) denotes an overlap area of vehicle collision determination regions at the predicted moment t′, As(t′) denotes an overlap area of vehicle safety reservation regions at the predicted moment t′, ω11 and ω12 denote a collision weight and a safety reservation weight, respectively, and I(Ac) and I(As) denote indicator functions, I(Ac) takes a value of 1 when the vehicle collision determination regions overlap and takes a value of 0 when the vehicle collision determination regions do not overlap, I(As) takes a value of 1 when the vehicle safety reservation regions overlap and takes a value of 0 when the vehicle safety reservation regions do not overlap;
(2) the time prediction payoff, denoted as:
Term(time)=v(t′)
wherein, v(t′) denotes a driving speed of the rear vehicle in the target lane in the game at the predicted moment t′;
(3) the comfort prediction payoff:
taking a derivative, denoted as Jerk, of an acceleration of the specified vehicle or an acceleration of the rear vehicle in the target lane at the predicted moment t′ during a vehicle driving process as the comfort prediction payoff, denoted as:
Term(cf)=−|Jerk(t′)|
wherein, Jerk(t′) denotes the derivative of the acceleration of the specified vehicle or the acceleration of the rear vehicle in the target lane at the predicted moment t′; and
(4) the cooperation prediction payoff:
taking an acceleration aj(t′) of the rear vehicle in the target lane in the game at the predicted moment t′ as a quantitative index of the cooperation prediction payoff, denoted as:
Term(gt)=−|aj(t′)|
wherein the payoff of the specified vehicle and the payoff of the rear vehicle in the target lane are determined by combining and weighting, respectively:

OG Complex Work Unit Math
wherein, ω=[ω1, ω2, ω3, ω4] and σ=[σ1, σ2, σ3, σ4], ω and σ denote weight coefficients,
wherein the driver aggressiveness factor βt is used to construct the weight coefficients of the payoff U of the specified vehicle:

OG Complex Work Unit Math
wherein k=[k1, k2, k3, k4], k1, k2, k3, and k4 denote gain factors for the safety prediction payoff, the time prediction payoff, the comfort prediction payoff, and the cooperation prediction payoff, respectively.