Autonomous Driving Maneuver Selection With Disallowed Action Filtering
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing semi-autonomous driving systems face challenges in preventing disallowed driving maneuvers, which can be selected due to the generalization performance of machine learning models, leading to potential safety risks and the need for frequent retraining when traffic rules change.
Innovation Solution
A method that uses sensor data to create a situation representation mapped to a probability distribution, where disallowed driving maneuvers are set to zero in the probability distribution or penalized in the cost function, ensuring only allowable maneuvers are selected, and allows for independent enforcement of boundary conditions without affecting the machine learning model's training.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If a machine learning model is used to map situation representation to driving maneuvers, then the system can generalize to new situations, but disallowed maneuvers may still be selected with non-zero probability
Solution Approach 1:
The system segments the driving maneuver selection process into two independent parts: (1) the machine learning model that generates probability distributions based on situation representation, and (2) the boundary condition enforcement module that independently filters disallowed maneuvers. This segmentation allows the ML model to maintain its generalization capability while the boundary conditions ensure safety by preventing disallowed maneuvers regardless of the probability distribution.
Solution Approach 2:
The patent introduces an intermediary component (the boundary condition enforcement mechanism) that sits between the machine learning model and the final maneuver execution. This intermediary checks whether selected maneuvers satisfy boundary conditions and can override the ML model's probabilistic selections, thus mediating between the model's generalization performance and safety requirements.
2Reliability
If boundary conditions are included in the machine learning model training, then disallowed maneuvers can be prevented, but the model's generalization performance decreases
Solution Approach 1:
The system separates boundary condition enforcement from the machine learning model training process. Boundary conditions are enforced independently after the ML model generates its probability distribution, rather than being incorporated into the model's learning objective. This maintains the model's generalization performance while ensuring disallowed maneuvers are prevented.
Solution Approach 2:
The system performs preliminary classification of maneuvers into allowed and disallowed categories based on boundary conditions before final selection. By pre-defining which maneuvers are permissible regardless of the ML model's output, the system ensures safety without constraining the model's learning and generalization capabilities.
3Adaptability or versatility
If traffic rules change, then new safety requirements must be enforced, but retraining the machine learning model is required
Solution Approach 1:
The system makes the boundary conditions dynamic and configurable without requiring model retraining. When traffic rules change, the boundary condition parameters can be updated independently to reflect new regulations, and the system adapts immediately without needing to retrain the machine learning model, thus avoiding time loss while maintaining adaptability.
Data Source
AI summary
A method for selecting a driving maneuver to be carried out by an at least semi-autonomously driving vehicle is disclosed. The method includes (i) using measurement data of at least one sensor carried by the vehicle, creating a representation of the situation the vehicle is in, (ii) mapping the representation of the situation to a probability distribution by way of a trained machine learning model, which probability distribution specifies a probability for every driving maneuver from a predefined catalog of available driving maneuvers, with which said driving maneuver is carried out, (iii) selecting a driving maneuver from the probability distribution as the driving maneuver to be carried out, (iv) in addition to using at least one aspect of the situation the vehicle is in, a subset of driving maneuvers which are disallowed in this situation is determined, and (v) this disallowed driving maneuver is prevented from being carried out.