US 12,293,663 B2
System and method for parking spot selection using reinforcement learning
Alireza Esna Ashari Esfahani, Novi, MI (US)
Assigned to GM GLOBAL TECHNOLOGY OPERATIONS LLC, Detroit, MI (US)
Filed by GM Global Technology Operations LLC, Detroit, MI (US)
Filed on Nov. 21, 2022, as Appl. No. 18/057,274.
Prior Publication US 2024/0169839 A1, May 23, 2024
Int. Cl. G08G 1/14 (2006.01); G01C 21/34 (2006.01); G06N 3/092 (2023.01); G06V 20/58 (2022.01); H04W 4/40 (2018.01)
CPC G08G 1/141 (2013.01) [G06N 3/092 (2023.01); G01C 21/3476 (2013.01); G06V 20/584 (2022.01); H04W 4/40 (2018.02)] 18 Claims
OG exemplary drawing
 
1. A method to select a parking spot using reinforcement learning, comprising:
loading a map of a parking lot into a host vehicle, the parking lot map including designated parking spots;
loading an initial belief of a parking spot occupancy of each of the designated parking spots, wherein the initial belief is an initial believed probability that each of the designated parking spots is occupied;
identifying that a parking lot communication is available to the host vehicle when one of a (vehicle-to-infrastructure) V2I communication, a (vehicle-to-vehicle) V2V communication and a V2X (vehicle-to-anything) communication is available and a parking lot operator command is available;
updating the initial belief to a belief matrix based on available parking spots communicated from the V2I and the V2V;
retrieving information from a perception program including at least one of a host vehicle camera and a camera in a parking infrastructure, and further updating the belief matrix;
receiving a user preference input that corresponds to preferences for desired characteristics of the designated parking spots;
running reinforcement learning (RL) to select one of the designated parking spots based on the user preference input and the belief matrix; and
commanding the host vehicle to move to the selected one of the designated parking spots.