	US 12,205,004 B2
	Systems and methods for training probabilistic object motion prediction models using non-differentiable prior knowledge
Sergio Casas, Toronto (CA); Cole Christian Gulino, Pittsburgh, PA (US); Shun Da Suo, Toronto (CA); and Raquel Urtasun, Toronto (CA)
Assigned to AURORA OPERATIONS, INC., Pittsburgh, PA (US)
Filed by Aurora Operations, Inc., Pittsburgh, PA (US)
Filed on Oct. 26, 2023, as Appl. No. 18/495,434.
Application 18/495,434 is a continuation of application No. 17/150,798, filed on Jan. 15, 2021, granted, now 11,836,585.
Claims priority of provisional application 63/123,251, filed on Dec. 9, 2020.
Claims priority of provisional application 62/984,034, filed on Mar. 2, 2020.
Prior Publication US 2024/0054407 A1, Feb. 15, 2024
Int. Cl. G06N 20/00 (2019.01); G05B 13/02 (2006.01); G05D 1/00 (2006.01); G06N 7/01 (2023.01)

CPC G06N 20/00 (2019.01) [G05B 13/0265 (2013.01); G05D 1/0088 (2013.01); G06N 7/01 (2023.01)]

20 Claims

1. A computer-implemented method for autonomous vehicle motion control, the method comprising:

obtaining sensor data for an environment comprising an object;

processing the sensor data with a machine-learned object motion prediction model to obtain a predicted location probability distribution for a future location of the object at one or more future times, wherein the machine-learned object motion prediction model has been trained by performing a REINFORCE gradient estimation technique to determine an approximate gradient of an expected loss that is a function of a non-differentiable prior knowledge reward function that encodes prior knowledge about motion of the object;

generating a vehicle trajectory based, at least in part, on the predicted location probability distribution for the future location of the object at the one or more future times; and

controlling motion of an autonomous vehicle based on the generated vehicle trajectory.