Vehicle Door State Detection Using ML Sensor Fusion
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing computer vision techniques for autonomous vehicles may not accurately detect certain door states of parked vehicles, leading to potential unsafe decisions in high-risk environments.
Innovation Solution
The use of machine-learned models to detect and predict door states of vehicles based on sensor data, allowing for accurate determination of door states such as open, closed, opening, or closing, and enabling informed decision-making for autonomous vehicle operation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If existing computer vision techniques are used to detect door states, then the system is simple and easy to implement, but the detection accuracy is insufficient leading to unsafe decisions
Solution Approach 1:
The patent replaces traditional computer vision techniques with machine-learned models that process sensor data. This substitution enables more accurate door state detection by using trained models that can distinguish between open, closed, opening, and closing states, thereby resolving the contradiction between detection accuracy and system complexity.
Solution Approach 2:
The patent changes the detection parameters by using multiple sensor data types and processing them through machine-learned models. This allows the system to accurately detect door states that were previously undetectable or misclassified, improving measurement precision while managing complexity through automated model-based processing.
2Reliability
If traditional sensor data processing is used, then the processing is fast and simple, but the ability to detect subtle door states is insufficient
Solution Approach 1:
The patent replaces traditional sensor data processing methods with machine-learned models that can interpret subtle patterns in sensor data. This substitution improves reliability by enabling accurate detection of door states such as opening and closing, which are difficult to detect using conventional processing methods.
3Measurement precision
If machine-learned models are implemented to improve door state detection, then detection accuracy improves, but computational requirements and processing time increase
Solution Approach 1:
The patent applies preliminary action by training machine-learned models in advance to recognize door states. Once trained, these models can quickly process sensor data during actual operation, achieving high detection accuracy without significant processing delays during critical moments.
Data Source
AI summary
Techniques for detecting door states associated with vehicles in an environment are described herein. The techniques may include receiving sensor data associated with a first vehicle in an environment and inputting the sensor data into a machine-learned model that is configured to determine a state of a door of the first vehicle. Based on the input data, an output may be received from the machine-learned model that indicates the state of the door of the first vehicle. Based on the state of the door of the first vehicle, a trajectory of a second vehicle may be controlled. The techniques may also include training the machine-learned model to detect door states of vehicles.