AV Trajectory Planning With Distilled Neural Prediction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The integration of machine learning models with optimization-based systems for solving constrained optimization problems in safety-critical tasks, such as autonomous vehicle trajectory planning, is challenging due to the non-convexity and complexity of machine learning models, leading to increased computational complexity and reliance on heuristic methods.
Innovation Solution
The use of knowledge distillation to train a smaller student prediction network based on a larger teacher prediction network, enabling efficient and accurate prediction of autonomous vehicle trajectories while satisfying safety constraints, thereby facilitating integration with classical optimization-based systems.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If machine learning models are used for trajectory planning, then accuracy and efficiency are improved, but computational complexity and difficulty in satisfying safety constraints increase
Solution Approach 1:
The system segments the trajectory planning problem into two distinct components: a machine learning-based prediction network that handles accurate trajectory prediction, and a separate optimization module that handles constraint satisfaction and safety guarantees. This segmentation allows each component to specialize in its strength while avoiding the weaknesses of the other.
Solution Approach 2:
The patent introduces an intermediary optimization framework that acts as a bridge between the machine learning prediction model and the safety constraints. This intermediary layer transforms the complex non-convex optimization problem into a more manageable form that can be solved efficiently while maintaining safety guarantees.
2Reliability
If machine learning models are integrated with optimization-based systems, then safety constraint satisfaction is improved, but solution feasibility and real-time performance deteriorate
Solution Approach 1:
The system performs preliminary actions by pre-training the prediction network on extensive data and pre-computing optimization parameters offline. This preliminary preparation enables the online system to quickly solve optimization problems in real-time without compromising safety constraints, as the heavy computational lifting has already been done in advance.
Solution Approach 2:
The patent employs parameter changes by transforming the optimization variables and constraints into a parameterized form that can be efficiently solved. By changing the parameterization of the trajectory and constraint representations, the system achieves both real-time performance and safety guarantee satisfaction.
3Measurement precision
If larger prediction networks are used, then prediction accuracy is improved, but computational efficiency and scalability worsen
Solution Approach 1:
The patent extracts the essential predictive capabilities from large, complex prediction networks and isolates them into a streamlined model that retains accuracy for trajectory prediction tasks. By taking out only the necessary predictive functions and removing redundant computational elements, the system achieves both accuracy and efficiency.
Data Source
AI summary
An electronic device and a method for AV trajectory planning using neural networks trained based on knowledge distillation is provided. A set of updated values of a set of variables of an objective function for trajectory planning of an ego AV are determined. A first prediction network is applied on an updated value of the set of updated values, a states of the ego AV and a set of AVs over a past time interval. Based on the application, an output is determined. The output includes states of the ego AV and set of AVs over a future time interval. The electronic device determines a set of optimal values based on the updated value and the determined output satisfying a safety constraint associated with the objective function. Further, the electronic device controls a trajectory of the ego AV based on the set of optimal values of the set of variables.