Vehicle Path Cost Volume Using Driving Behavior Adjustments
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing path planning systems for autonomous vehicles face limitations in handling non-edge cases and scalability, requiring expert knowledge and being time-consuming, while relying solely on observed driving behavior lacks interpretability and flexibility.
Innovation Solution
Integrate hand-engineered costs with machine-learning models to generate a cost volume that incorporates both movement restrictions and observed driving behavior, using convolutional neural networks to adjust initial cost volumes based on environment data.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of operation
If hand-engineered costs are used to generate cost volume, then interpretability and flexibility are improved, but scalability and handling of non-edge cases deteriorate
Solution Approach 1:
The patent merges hand-engineered cost functions with machine-learning-derived cost functions into a unified cost volume. The hand-engineered costs provide interpretability and flexibility for edge cases, while the machine-learning component handles non-edge cases and scales to diverse scenarios. This combination allows the system to leverage both the transparency of rule-based approaches and the adaptability of data-driven approaches.
2Adaptability or versatility
If machine-learning models are used to generate cost volume from observed driving behavior, then scalability is improved, but interpretability and flexibility deteriorate
Solution Approach 1:
The patent introduces hand-engineered cost functions as an intermediary layer between the machine-learning model and the final trajectory selection. This intermediary provides interpretability by offering explicit, human-understandable cost criteria that can be inspected and adjusted. The machine-learning model generates base costs from observed driving behavior, which are then refined and explained through the hand-engineered cost functions.
3Reliability
If trajectory optimization is used to find potential vehicle-trajectory, then safety and feasibility are improved, but computational time and complexity deteriorate
Solution Approach 1:
The patent performs preliminary action by pre-computing cost volumes for multiple potential trajectories before actual trajectory optimization. The cost volume encodes safety, feasibility, and comfort information in advance, allowing the optimization process to efficiently evaluate and compare pre-scored trajectories rather than computing all constraints from scratch during real-time optimization.
Data Source
Figure 1A~1B
Figure 1C
Figure 1D~2
AI summary
In one embodiment, a method includes determining an initial cost volume associated with a plurality of potential trajectories of a vehicle in an environment based on a set of movement restrictions of the vehicle, generating a delta cost volume using the initial cost volume and environment data associated with the environment, wherein the delta cost volume is generated by determining adjustments to the initial cost volume that incorporate observed driving behavior, and scoring a trajectory of the plurality of potential trajectories for the vehicle based on t the initial cost volume and the delta cost volume.