Yielding Prediction Model for Autonomous Vehicle Agent Interaction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current autonomous vehicle systems face challenges in accurately predicting yielding behaviors of agents with overlapping future paths, due to noisy training data and difficulty in selecting appropriate features for machine learning models, which affects the generation of timely and safe planning decisions.
Innovation Solution
An on-board system that uses a yielding prediction model, trained on focused driving scenarios where the target agent has lower or equal precedence, to generate likelihood scores for agent interactions, incorporating features that describe both agents and their predicted paths, enabling the system to identify which agent will yield and modify trajectory predictions accordingly.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a machine learning model is used to predict yielding behaviors of agents with overlapping future paths, then the accuracy of predicting which agent will yield improves, but the complexity of selecting appropriate features and training data increases
Solution Approach 1:
The patent segments the feature set into distinct categories: agent features (position, velocity, acceleration, heading), path features (curvature, length, time-to-collision), and interaction features (relative position, relative velocity, time-to-Conflict). This segmentation makes feature selection more manageable and systematic, resolving the complexity issue while maintaining prediction accuracy.
Solution Approach 2:
The patent performs preliminary action by pre-defining and organizing feature templates before training the machine learning model. The system prepares structured feature representations including agent dynamics, path geometry, and interaction metrics in advance, which simplifies the training process and improves prediction accuracy without requiring complex feature engineering during model execution.
2Adaptability or versatility
If training data is collected from real-world driving scenarios, then the model learns from actual agent behaviors, but the noisy and inconsistent nature of real-world data reduces prediction reliability
Solution Approach 1:
The patent applies beforehand cushioning by implementing data filtering and preprocessing steps before training the model. The system filters out noisy or inconsistent real-world data through validation rules and preprocessing techniques, cushioning the model training process from the adverse effects of poor-quality data while still maintaining adaptability to real-world driving scenarios.
Solution Approach 2:
The patent introduces an intermediary layer between raw real-world data and the machine learning model. This intermediary includes data preprocessing, feature engineering, and validation mechanisms that mediate the transformation of noisy real-world data into clean, structured training data, thereby improving prediction reliability while preserving model adaptability.
3Reliability
If the system predicts yielding likelihood for multiple agent pairs with overlapping paths, then the comprehensiveness of safety assessment improves, but the computational time and processing resources increase
Solution Approach 1:
The patent applies partial action by prioritizing the prediction of yielding likelihood for agent pairs with higher conflict risk. The system calculates yielding probabilities for multiple agent pairs but focuses computational resources on those with smaller time-to-collision, higher velocity, or more constrained paths, achieving comprehensive safety assessment without processing all possible pairs equally, thus reducing computational time.
4Loss of information
If the machine learning model outputs detailed likelihood scores for each agent pair, then the planning subsystem can make more informed decisions, but the quantity of data to be processed increases
Solution Approach 1:
The patent extracts only the most critical information from the machine learning model outputs for the planning subsystem. Instead of processing all detailed likelihood scores, the system extracts key metrics such as the highest yielding probability, critical time-to-conflict values, and identified yielding agents, reducing data processing complexity while maintaining information completeness for safe planning decisions.
Data Source
AI summary
Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, for predicting how likely it is that a target agent in an environment will yield to another agent when the pair of agents are predicted to have overlapping future paths. In one aspect, a method comprises obtaining a first trajectory prediction specifying a predicted future path for a target agent in an environment; obtaining a second trajectory prediction specifying a predicted future path for another agent in the environment; determining that, at an overlapping region, the predicted future path for the target agent overlaps with the predicted future path for the other agent; and in response: providing as input to a machine learning model respective features for the target agent and the other agent; and obtaining the likelihood score as output from the machine learning model.