Autonomous Vehicle Trajectory Prediction With Uncertainty Modeling
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing machine learning-based systems for autonomous vehicles face inaccuracies in trajectory predictions due to uncertainties, leading to potential unsafe situations, as they do not fully account for variations in vehicle states and environmental interactions.
Innovation Solution
The development of techniques that train machine learning models to generate predicted trajectories and associated uncertainties, using a combination of coarse-step and fine-step prediction methods, along with backpropagation techniques, to improve accuracy and reduce computational resources, allowing for safer and more flexible vehicle operation by accounting for multiple possible paths within uncertainty bounds.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If machine learning models are used for trajectory prediction in autonomous vehicles, then the system can detect objects and predict paths, but inaccuracies in predictions and uncertainties create unsafe situations
Solution Approach 1:
The patent segments the trajectory prediction process into multiple discrete time steps, where the model predicts a sequence of future states step-by-step rather than attempting to predict the entire trajectory in one step. This segmentation allows the system to manage uncertainty at each individual step while maintaining overall prediction accuracy throughout the trajectory.
Solution Approach 2:
The patent performs preliminary actions by training the machine learning model to explicitly predict uncertainty alongside trajectory predictions during the training phase. The model learns to output both predicted states and associated uncertainties before actual deployment, enabling the system to prepare for potential inaccuracies in advance and make more reliable safety decisions.
2Reliability
If the system accounts for uncertainties in trajectory predictions, then safety improves, but computational resources increase
Solution Approach 1:
The patent applies partial action by having the model predict uncertainty only for the most critical aspects of the trajectory at each time step, rather than computing exhaustive uncertainty for every possible parameter. This selective approach provides sufficient safety information while keeping computational demands manageable for real-time autonomous vehicle operation.
3Adaptability or versatility
If the model predicts multiple possible paths within uncertainty bounds, then flexibility and safety improve, but the complexity of the system increases
Solution Approach 1:
The patent makes the machine learning model universal by designing it to simultaneously perform multiple functions: predicting the trajectory, calculating uncertainties, and generating multiple possible paths within uncertainty bounds. This multi-functional approach increases adaptability without requiring separate dedicated systems for each function, thereby managing complexity while enhancing versatility.
Data Source
AI summary
A remote computer system may receive data representative of a state of an autonomous vehicle traversing an environment. A representation of the environment and the state may be input to a first machine learned model, which may output a vector representation of the environment at a first time. The vector representation may be input to a second machine learned model, which may output a second vector representation of the environment at a second time. A third machine learned model may receive the second vector representation and may output a mean velocity and a mean steering angle of the vehicle at the first time and associated uncertainty variables. Based on the mean velocity, the mean steering angle, and the uncertainty variables, a predicted vehicle state at the first time and an associated uncertainty may be determined and the vehicle may be controlled based on the predicted vehicle state and the uncertainty.