Corner case detection and collection for a path planning system
AssigneesVolkswagen Aktiengesellschaft, Audi AG, Porsche AG
InventorsXinwei Gong, Sai Prasad Nooka
Life of this patent
Estimating the term…
Abstract
In a predictive path planning functionality for a transportation vehicle data describing a roadway are acquired by one or more sensors disposed on the transportation vehicle and analyzed. One or more predicted paths are predicted with at least one neural network. Data describing a reference path taken by the transportation vehicle are acquired. The one or more predicted paths are evaluated to detect one or more errors in the predicting operation wherein the one or more errors include an indication that the one or more predicted paths of the path planning system deviate from the reference path.
Claims22
Transportation vehicle equipment for detecting errors in a path predicting neural network, the equipment being disposed on the transportation vehicle and comprising:
- one or more sensors for acquiring data describing a roadway and a reference path taken by the transportation vehicle;
- a processor and memory comprising a ring buffer for continuously storing data from the one or more sensors and a semi-permanent memory;
- at least one neural network for predicting one or more predicted paths for the transportation vehicle;
- wherein the processor is configured as a means for evaluating the one or more predicted paths to detect one or more errors in a predicting operation wherein the one or more errors include an indication that the one or more predicted paths of a path planning system deviate from the reference path;
- and
- wherein, responsive to detection of the one or more errors in the predicting operation, data are transmitted from the ring buffer to the semi-permanent memory and wherein the data transmitted from the ring buffer to the semi-permanent memory includes a first portion of data corresponding to a period of time before the detected error and a portion of data corresponding to a period of time after the detected error.
A method of error detection for a path predicting neural network, the method comprising:
- analyzing data describing a roadway acquired by one or more sensors disposed on a transportation vehicle;
- predicting with at least one neural network, disposed on a transportation vehicle, one or more predicted paths for the transportation vehicle;
- acquiring data describing a reference path taken by the transportation vehicle;
- evaluating the one or more predicted paths to detect one or more errors in the predicting operation wherein the one or more errors include an indication that the one or more predicted paths of the path planning system deviate from the reference path;
- and
- transmitting data, responsive to detection of the one or more errors in the predicting operation, from a ring buffer for continuous storage of data of the one or more sensors to a semi-permanent memory, wherein the data transmitted from the ring buffer to the semi-permanent memory includes a first portion of data corresponding to a period of time before the detected error and a portion of data corresponding to a period of time after the detected error.
A computing system for planning a path of a transportation vehicle, the system disposed on the transportation vehicle and comprising:
- at least one memory configured to receive sensor data indicative of a reference path for the vehicle;
- at least one processor communicatively coupled to the at least one memory, wherein the at least one processor is configured to execute program instructions to cause the system to perform the steps comprising:
- predicting one or more predicted paths for the transportation vehicle by at least one neural network;
- detecting a deviation from the one or more predicted paths;
- determining a corner case based on the reference path and the detected deviation;
- storing the corner case, wherein storing includes transmitting data, responsive to detection of the one or more errors in the predicting operation, from a ring buffer of the at least one memory for continuous storage of data of the one or more sensors to a semi-permanent memory of the at least one memory, wherein the data transmitted from the ring buffer to the semi-permanent memory includes a first portion of data corresponding to a period of time before the detected error and a portion of data corresponding to a period of time after the detected error;
- and
- training the at least one neural network with the stored corner case.
Dependencies are inferred from the claim text.