Map-Based Scenario Annotation for Autonomous Path Training
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Autonomous vehicle navigation systems face challenges in providing accurate self-navigation decisions due to incomplete perception and reliability issues with neural networks, and existing training datasets are inadequate for handling all possible scenarios, leading to misbehavior in unknown environments.
Innovation Solution
The method involves augmenting training sets with synthetically modified scenarios, including path altering and non-altering objects, to enhance the autonomous driving model's ability to navigate by modifying existing driving scenarios with synthetic objects, thereby improving the model's decision-making capabilities and adaptability to new environments.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If existing training datasets are used for autonomous driving models, then the model can be trained with available data, but the model exhibits misbehavior in unknown environments due to inadequate scenario coverage
Solution Approach 1:
The patent applies preliminary action by pre-modifying training scenarios to include diverse synthetic objects and path alterations before the model is deployed. This prepares the model in advance for unknown environmental conditions by exposing it to varied synthetic scenarios during training, thereby improving adaptability while maintaining reliability through comprehensive pre-training coverage
Solution Approach 2:
The patent employs parameter changes by systematically varying scenario parameters such as object types, positions, and path alterations in the training data. By changing these parameters to create diverse synthetic scenarios, the model learns to handle a broader range of conditions, improving both adaptability to unknown environments and reliability of decisions
2Adaptability or versatility
If manual algorithm development is used to handle all possible scenarios, then comprehensive scenario coverage can be achieved, but the system complexity and development time increase significantly
Solution Approach 1:
The patent applies copying by creating synthetic copies of real driving scenarios with modified parameters. Instead of manually developing algorithms for each possible scenario, the system copies existing scenarios and introduces synthetic variations, thereby achieving comprehensive scenario coverage while significantly reducing development complexity
Solution Approach 2:
The patent implements universality by creating a unified training framework that handles diverse scenarios through a single system. The synthetic object introduction mechanism and automated scenario modification approach provide multi-functional capability to cover various driving conditions without requiring separate manual algorithms for each scenario type
Data Source
AI summary
A method, an apparatus and a computer program product for enhancing training of autonomous movement models based on map-based annotations. The method comprises augmenting an original training set with map based annotated modified scenarios at least one of which affects a path and at least one of which does not affect a path. The method comprises generating modified driving scenarios introducing a synthetic path altering object into the functional map representing the original movement or driving scenario, that induces a driving path that is different than the original path; and other modified driving scenarios that introduce a path non-altering object into the functional map of the original scenario, that the original path is applicable therein. An autonomous driving model for predicting driving path within a road segment based on a functional map representation is trained using the modified driving scenarios.