18358675. GENERATING AND/OR USING TRAINING INSTANCES THAT INCLUDE PREVIOUSLY CAPTURED ROBOT VISION DATA AND DRIVABILITY LABELS simplified abstract (Google LLC)

GENERATING AND/OR USING TRAINING INSTANCES THAT INCLUDE PREVIOUSLY CAPTURED ROBOT VISION DATA AND DRIVABILITY LABELS

Organization Name

Google LLC

Inventor(s)

Ammar Husain of San Francisco CA (US)

Joerg Mueller of Mountain View CA (US)

GENERATING AND/OR USING TRAINING INSTANCES THAT INCLUDE PREVIOUSLY CAPTURED ROBOT VISION DATA AND DRIVABILITY LABELS - A simplified explanation of the abstract

This abstract first appeared for US patent application 18358675 titled 'GENERATING AND/OR USING TRAINING INSTANCES THAT INCLUDE PREVIOUSLY CAPTURED ROBOT VISION DATA AND DRIVABILITY LABELS

Simplified Explanation

Implementations described in this patent application involve generating training data for machine learning models used by robots to determine drivability of areas captured in vision data. The training data includes vision data and corresponding drivability labels. The drivability labels are determined using first vision data from a first vision component connected to the robot and processed using geometric and/or heuristic methods.

• Training data is generated for machine learning models used by robots.
• Each instance of training data includes vision data and drivability labels.
• Drivability labels are determined using first vision data from a first vision component.
• Drivability labels are generated by processing the first vision data using geometric and/or heuristic methods.
• Second vision data is generated using a second vision component, such as a camera.
• Drivability labels are correlated to the second vision data.
• The trained models can be shared with robots to enable them to determine drivability of areas captured in real-time vision data.

Potential Applications

• Autonomous driving: Robots can use the trained models to determine the drivability of different areas while navigating autonomously.
• Surveillance systems: Robots equipped with vision components can analyze the drivability of areas captured in real-time vision data for security purposes.
• Industrial automation: Robots can assess the drivability of different areas in manufacturing facilities to optimize operations and ensure safety.

Problems Solved

• Lack of efficient training data generation for machine learning models used by robots.
• Difficulty in determining drivability labels for vision data captured by robots.
• Limited ability of robots to assess the drivability of areas in real-time vision data.

Benefits

• Improved accuracy: The use of machine learning models trained with accurate drivability labels enhances the ability of robots to determine the drivability of areas captured in vision data.
• Real-time analysis: Robots can assess the drivability of areas in real-time vision data, allowing for immediate decision-making and action.
• Versatility: The trained models can be shared with multiple robots, enabling them to perform drivability assessments across various applications and environments.

Original Abstract Submitted

Implementations set forth herein relate to generating training data, such that each instance of training data includes a corresponding instance of vision data and drivability label(s) for the instance of vision data. A drivability label can be determined using first vision data from a first vision component that is connected to the robot. The drivability label(s) can be generated by processing the first vision data using geometric and/or heuristic methods. Second vision data can be generated using a second vision component of the robot, such as a camera that is connected to the robot. The drivability labels can be correlated to the second vision data and thereafter used to train one or more machine learning models. The trained models can be shared with a robot(s) in furtherance of enabling the robot(s) to determine drivability of areas captured in vision data, which is being collected in real-time using one or more vision components.