US 11,731,651 B2
Automatic parameter tuning framework for controllers used in autonomous driving vehicles
Weiman Lin, Sunnyvale, CA (US); Yu Cao, Sunnyvale, CA (US); Yu Wang, Sunnyvale, CA (US); Qi Luo, Sunnyvale, CA (US); Shu Jiang, Sunnyvale, CA (US); Xiangquan Xiao, Sunnyvale, CA (US); Longtao Lin, Sunnyvale, CA (US); Jinghao Miao, Sunnyvale, CA (US); and Jiangtao Hu, Sunnyvale, CA (US)
Assigned to BAIDU USA LLC, Sunnyvale, CA (US)
Filed by Baidu USA LLC, Sunnyvale, CA (US)
Filed on Sep. 30, 2020, as Appl. No. 17/39,685.
Prior Publication US 2022/0097728 A1, Mar. 31, 2022
Int. Cl. B60W 10/04 (2006.01); B60W 60/00 (2020.01); G01C 21/34 (2006.01); B60W 30/18 (2012.01); B60W 10/18 (2012.01); B60W 10/20 (2006.01)
CPC B60W 60/0011 (2020.02) [B60W 10/04 (2013.01); B60W 10/18 (2013.01); B60W 10/20 (2013.01); B60W 30/18009 (2013.01); G01C 21/3453 (2013.01)] 20 Claims
OG exemplary drawing
 
1. A computer-implemented method of automatic tuning of one or more controllers of an autonomous driving vehicle (ADV) control system, comprising:
for each of a plurality of driving scenarios:
running a simulation of the driving scenario using a first value for each of a plurality of tunable parameters comprising a plurality of lateral dynamics parameters in a state weighting matrix of a linear quadratic regulator (LQR) of an ADV to control one or more controllers of the ADV, the plurality of tunable parameters also comprising a time constant parameter and an adaptive gain parameter for a model-referenced adaptive gain controller; and
generating a score of the simulation of the driving scenario indicating performance of the one or more controllers during the simulation of the driving scenario;
computing a first weighted score from the scores generated for the simulation of each of the plurality of driving scenarios; and
optimizing the values of the plurality of tunable parameters by iteratively performing operations of running the simulation, generating the score, and computing the first weighted score, using the first weighted score for each repetition as an objective to generate a second value for each of the plurality of tunable parameters, wherein the optimized values for the plurality of tunable parameters are utilized by the one of more controllers of the ADV to navigate the ADV, and wherein the iterative operations are repeated until a second weighted score differs from the first weighted score of a previous iteration by less than a predetermined threshold amount or a predetermined fixed number of iterations has been performed.