Predicting Three-Dimensional Features For Autonomous Driving
Patent No. US12014553 (titled "Predicting Three-Dimensional Features For Autonomous Driving") was filed by Tesla Inc on Oct 14, 2021.
What is this patent about?
’553 is related to the field of autonomous vehicle control systems, specifically those employing machine learning. These systems rely on sensor data, such as camera images, to perceive the environment and make driving decisions. A key challenge is creating accurate and robust training data for the machine learning models, especially for predicting complex features like lane lines and the paths of other vehicles.
The underlying idea behind ’553 is to improve the accuracy of machine learning models used in autonomous driving by creating training data that leverages a time series of sensor data . Instead of relying on single snapshots, the system analyzes a sequence of images and sensor readings to determine a more accurate "ground truth" for features in the environment. This ground truth is then used to train the model to predict these features from a single image, enabling more robust and reliable autonomous driving.
The claims of ’553 focus on a system, method, and computer program product that obtain sensor data from a vehicle, determine a three-dimensional feature associated with the sensor data using a machine learning model, and adjust the vehicle's operation based on this feature. The machine learning model is trained using a dataset comprising a ground truth and corresponding sensor data captured over a period of time, where the model predicts the ground truth from a single time series element.
In practice, the system captures a video sequence and odometry data as the vehicle moves. By analyzing this sequence, the system can identify lane lines, even when partially occluded or poorly visible in individual frames. The system then constructs a 3D representation of the lane line based on the most accurate data from the entire sequence. This 3D representation, the ground truth, is then paired with a single image from the sequence to create training data. The trained model can then predict the 3D lane geometry from a single image.
This approach differs from traditional methods that rely on manual annotation of individual images or simpler 2D feature extraction. By using a time series to establish a more accurate ground truth, the system can train models that are more robust to noise, occlusion, and variations in lighting and weather. The ability to predict 3D trajectories also allows for more precise lane keeping and path planning, improving the safety and reliability of autonomous driving.
How does this patent fit in bigger picture?
Technical landscape at the time
In the late 2010s when ’553 was filed, autonomous driving systems were at a time when machine learning models were being increasingly used for perception tasks such as lane detection and object recognition, at a time when significant effort was being directed towards improving the quality and accuracy of training datasets for these models, and at a time when systems commonly relied on sensor data from cameras, lidar, and radar to capture the environment around a vehicle.
Novelty and Inventive Step
Claims were rejected for indefiniteness under 35 U.S.C. 112(b) and for obviousness under 35 U.S.C. 103. The rejection under 35 U.S.C. 103 was based on a combination of Kwant et al. and Song et al. The examiner indicated that certain claims would be allowable if rewritten to overcome the rejections under 35 U.S.C. 112(b) and to include all limitations of the base claim and any intervening claims. The prosecution record does NOT describe the technical reasoning or specific claim changes that led to allowance.
Claims
This patent contains 18 claims, of which claims 1, 10, and 18 are independent. The independent claims focus on a system, a method, and a computer program product, respectively, all relating to adjusting vehicle operation based on a three-dimensional feature determined from sensor data using a machine learning model. The dependent claims generally elaborate on and refine the elements and steps recited in the independent claims.
Key Claim Terms New
Definitions of key terms used in the patent claims.
Litigation Cases New
US Latest litigation cases involving this patent.
Patent Family
File Wrapper
The dossier documents provide a comprehensive record of the patent's prosecution history - including filings, correspondence, and decisions made by patent offices - and are crucial for understanding the patent's legal journey and any challenges it may have faced during examination.
Date
Description
Get instant alerts for new documents
US12014553
- Application Number
- US17450914
- Filing Date
- Oct 14, 2021
- Status
- Granted
- Expiry Date
- Dec 15, 2039
- External Links
- Slate, USPTO, Google Patents