US 12,403,922 B2
Method for assisting with driving an autonomous motor vehicle on a road
Hussam Atoui, Guyancourt (FR); David Gonzalez Bautista, Saint Cyr l'ecole (FR); Imane Mahtout, Guyancourt (FR); and Vicente Milanes, Valladolid (ES)
Assigned to AMPERE S.A.S., Boulogne Billancourt (FR)
Appl. No. 18/258,087
Filed by RENAULT S.A.S, Boulogne Billancourt (FR)
PCT Filed Dec. 7, 2021, PCT No. PCT/EP2021/084647
§ 371(c)(1), (2) Date Jun. 16, 2023,
PCT Pub. No. WO2022/128643, PCT Pub. Date Jun. 23, 2022.
Claims priority of application No. 20 13417 (FR), filed on Dec. 17, 2020.
Prior Publication US 2024/0034340 A1, Feb. 1, 2024
Int. Cl. B60W 50/06 (2006.01); B60W 50/00 (2006.01); B60W 60/00 (2020.01)
CPC B60W 50/06 (2013.01) [B60W 60/001 (2020.02); B60W 2050/0014 (2013.01); B60W 2530/10 (2013.01); B60W 2530/20 (2013.01); B60W 2552/05 (2020.02); B60W 2555/20 (2020.02)] 12 Claims
OG exemplary drawing
 
8. A device for assisting in driving of an autonomous motor vehicle on a road, said device comprising:
a fuzzy system comprising:
a fuzzying block configured to receive and process a plurality of data in order to determine a current situation of the autonomous motor vehicle;
an inference engine configured to interrogate a database in order to obtain a control response to be made with respect to said current situation; and
a defuzzying block configured to command an actuator in order to control the autonomous motor vehicle, according to the control response to be made; and
a reinforcement learning agent, said agent being configured to receive the command sent to the actuator and to generate a new optimized response with respect to the current situation according to an item of quality information on a result of the command for the control of said autonomous motor vehicle,
wherein the reinforcement learning agent is configured to receive the plurality of data, identify possible actions based on the plurality of data, calculate a value of the quality information for each of the possible actions, and select an action from the possible actions with a highest value of the quality information, the selected action being the new optimized response.