Object Yaw Estimation Using Multi-Representation Motion Models
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing systems for autonomous vehicles and similar applications face challenges in accurately determining the yaw angle of objects in an environment, often relying on single representations that can lose data due to linear estimation, leading to reduced accuracy in object tracking and navigation.
Innovation Solution
A machine learned model is trained to output a yaw angle based on multiple representations of objects, including those generated by sensor data and Bayesian filtering, reintroducing errors minimized in estimation to provide a more accurate representation of the object's heading.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If linear estimation is used to determine yaw angle from multiple representations, then computational simplicity is maintained, but measurement precision deteriorates due to loss of information and reduced accuracy
Solution Approach 1:
The patent replaces the mechanical/mathematical linear estimation system with a machine learned model (neural network) that processes multiple representations and ground truth yaw data. This substitution allows the system to learn non-linear relationships and patterns that linear estimation cannot capture, thereby improving measurement precision while maintaining computational feasibility through the trained model's efficient inference
Solution Approach 2:
The patent changes the fundamental approach from linear parameter estimation to non-linear machine learning parameter optimization. By training the model on ground truth yaw angles and multiple representations, the system learns optimal parameter transformations that preserve information and improve accuracy, moving away from the constrained linear estimation framework
2Productivity
If a single representation is used for object tracking, then processing speed is maintained, but measurement precision deteriorates due to data loss
Solution Approach 1:
The patent merges multiple representations (bounding boxes, orientations, positions from different sources) along with ground truth yaw data into a unified input for the machine learned model. This combination allows the system to leverage complementary information from each representation, improving measurement precision through data fusion while the model's efficient architecture maintains processing speed
Solution Approach 2:
The patent uses ground truth yaw angles as reference copies to train the machine learned model. By having access to accurate ground truth data during training, the model learns to produce accurate yaw estimates from multiple representations, effectively using the ground truth as a template for what the correct output should be
Data Source
AI summary
Techniques to use a trained model to determine a yaw of an object are described. For example, a system may implement various techniques to generate multiple representations for an object in an environment. Each representation vary based on the technique and data used. An estimation component may estimate a representation from the multiple representations. The model may be implemented to output a yaw for the object using the multiple representations, the estimated representation, and/or additional information. The output yaw may be used to track an object, generate a trajectory, or otherwise control a vehicle.