Machine-Learned Yield Decisions for Autonomous Vehicle Traffic Handling
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Autonomous vehicles face challenges in making effective yield decisions in complex traffic scenarios, relying on rule-based algorithms that are time-consuming and costly to develop and update.
Innovation Solution
A machine-learned yield model is implemented in an autonomous vehicle's computing system, processing feature data from perceived objects to provide yield decisions, thereby improving driving performance and reducing reliance on hand-crafted rules.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If rule-based algorithms are used for yield decisions, then the autonomous vehicle can make decisions based on explicit traffic rules, but the development and update of these rules is time-consuming and costly
Solution Approach 1:
The patent replaces the mechanical system of hand-crafted rule-based algorithms with a machine-learned model that automatically learns yield decision patterns from training data. This substitution eliminates the need for manual rule development and updates, allowing the system to adapt to new traffic scenarios through data-driven learning rather than manual programming.
Solution Approach 2:
The machine-learned yield model enables the autonomous vehicle to self-improve its yield decision-making capabilities by learning from training data without requiring external rule updates. The system serves itself by automatically adapting to new scenarios through continuous learning from observed traffic patterns and human driver behaviors.
2Productivity
If machine-learned yield model is implemented, then the autonomous vehicle can make faster and more adaptive yield decisions, but the model requires training data processing and computational resources
Solution Approach 1:
The patent applies preliminary action by pre-training the machine-learned yield model on extensive training data before deployment. This allows the model to learn complex yield decision patterns in advance, so that during actual operation, the vehicle can make rapid decisions without requiring complex real-time computations. The heavy computational work is performed beforehand during the training phase.
3Device complexity
If hand-crafted rules are used, then the system structure is simpler and easier to understand, but frequent rule updates are needed to handle complex traffic scenarios
Solution Approach 1:
The patent transforms the static parameter-based rule system into a dynamic model that adapts its parameters through learning. Instead of fixed rules that require manual updates, the machine-learned model continuously adjusts its internal parameters based on training data, enabling it to handle diverse traffic scenarios without changing its fundamental structure. The system maintains simplicity while gaining adaptability through parameter learning rather than rule modification.
Data Source
AI summary
The present disclosure provides autonomous vehicle systems and methods that include or otherwise leverage a machine-learned yield model. In particular, the machine-learned yield model can be trained or otherwise configured to receive and process feature data descriptive of objects perceived by the autonomous vehicle and/or the surrounding environment and, in response to receipt of the feature data, provide yield decisions for the autonomous vehicle relative to the objects. For example, a yield decision for a first object can describe a yield behavior for the autonomous vehicle relative to the first object (e.g., yield to the first object or do not yield to the first object). Example objects include traffic signals, additional vehicles, or other objects. The motion of the autonomous vehicle can be controlled in accordance with the yield decisions provided by the machine-learned yield model.