Autonomous Vehicle Yielding Using Pedestrian Crossing Intent Prediction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing autonomous and semi-autonomous vehicles lack the ability to accurately predict an agent's intent and action to cross a roadway, which can lead to unsafe operations and inefficient energy use.
Innovation Solution
A system utilizing machine learning models to generate crossing intent and action predictions based on scene features, allowing autonomous vehicles to determine future trajectories and speeds, and mimic human driver behavior.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If the autonomous vehicle yields to all potential crossing agents, then safety is improved, but energy consumption increases due to unnecessary yielding actions
Solution Approach 1:
The patent replaces traditional rule-based yielding logic with a machine learning model that predicts crossing intent and action. The system uses neural networks to process scene features and generate probability scores for crossing intent and crossing action, substituting mechanical decision-making with intelligent prediction-based control.
Solution Approach 2:
The system changes the decision parameter from binary yield/no-yield to probabilistic assessment. By calculating crossing intent probability and crossing action probability separately, the system can make nuanced decisions based on threshold comparisons, allowing it to distinguish between agents who will definitely cross and those who might only cross if the vehicle yields.
2Device complexity
If the autonomous vehicle uses traditional rule-based yielding, then decision-making is simple, but accuracy in predicting agent behavior is insufficient
Solution Approach 1:
The patent replaces simple rule-based decision-making with a sophisticated machine learning system. The encoder neural network processes multiple scene features including agent position, velocity, and environmental context, then passes them through intent prediction and action prediction neural networks to generate accurate probability assessments of crossing behavior.
Solution Approach 2:
The patent segments the behavior prediction task into two distinct components: crossing intent prediction and crossing action prediction. This segmentation allows the system to independently assess what the agent intends to do versus what the agent will actually do, providing more nuanced and accurate predictions than a single unified model.
3Measurement precision
If separate models are used for intent prediction and action prediction, then prediction accuracy is improved, but computing resource usage increases
Solution Approach 1:
The patent merges the intent prediction and action prediction into a single integrated machine learning model. Both predictions are generated simultaneously within one model architecture, sharing common encoding layers and processing scene features in a unified manner, which reduces redundant computations and improves efficiency.
Solution Approach 2:
The system performs preliminary encoding of scene features using a shared encoder neural network before branching into intent and action predictions. This preliminary processing extracts common representations once, avoiding redundant feature extraction and reducing overall computational burden while maintaining prediction accuracy.
Data Source
AI summary
Methods, systems, and apparatus, including computer programs encoded on a computer storage medium that determine yield behavior for an autonomous vehicle, and can include identifying an agent that is in a vicinity of an autonomous vehicle navigating through a scene at a current time point. Scene features can be obtained and can include features of (i) the agent and (ii) the autonomous vehicle. An input that can include the scene features can be processed using a first machine learning model that is configured to generate (i) a crossing intent prediction that includes a crossing intent score that represents a likelihood that the agent intends to cross a roadway in a future time window after the current time, and (ii) a crossing action prediction that includes a crossing action score that represents a likelihood that the agent will cross the roadway in the future time window after the current time.