Near-Curb Behavior Prediction for Multi-Agent Autonomous Driving
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing systems for predicting near-curb driving behavior in autonomous vehicles rely on hand-made heuristics, which are limited in accuracy and brittle, failing to account for various scenarios, especially those outside the designers' intentions, and can only generate predictions for a single agent at a single time point.
Innovation Solution
A machine learning model trained on real-world driving data from multiple vehicles is used to predict near-curb driving behavior for various agents, including vehicles, cyclists, and pedestrians, capable of generating predictions for multiple agents at multiple future time points, improving accuracy and adaptability.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of manufacture
If hand-made heuristics are used to predict near-curb driving behavior, then the system is simple to implement, but the prediction accuracy is limited and the system is brittle
Solution Approach 1:
The patent replaces hand-made heuristic rules with a machine learning model that processes sensor data to predict near-curb driving behavior. The machine learning model learns patterns from training data generated by vehicles operating in the real world, substituting manual rule-based systems with an adaptive computational approach that achieves higher prediction accuracy while maintaining automated operation.
2Device complexity
If hand-made heuristics are used, then the system design is straightforward, but the system fails to account for various scenarios outside designers' intentions
Solution Approach 1:
The patent employs a machine learning model that is trained in advance on extensive real-world driving data collected from multiple vehicles. This preliminary training enables the model to learn diverse driving scenarios and behaviors before deployment, allowing it to adapt to various situations including those outside the original designers' intentions, without requiring complex manual rule updates for each scenario.
3Productivity
If existing prediction systems are used, then they can only generate predictions for a single agent at a single time point, but using a machine learning model trained on real-world data enables predictions for multiple agents at multiple future time points
Solution Approach 1:
The patent extends the prediction capability from single-agent/single-timepoint to multi-agent/multi-timepoint predictions by utilizing a machine learning model that processes sequences of sensor data. The model predicts near-curb driving behavior for multiple agents across multiple future time points simultaneously, adding temporal and multi-object dimensions to the prediction output, which enables more comprehensive route planning and safety assessments.
Data Source
AI summary
Methods, systems, and apparatus, including computer programs encoded on computer storage media, for predicting near-curb driving behavior. One of the methods includes obtaining agent trajectory data for an agent in an environment, the agent trajectory data comprising a current location and current values for a predetermined set of motion parameters of the agent; processing a model input generated from the agent trajectory data using a trained machine learning model to generate a model output comprising a prediction of whether the agent will exhibit near-curb driving behavior within a predetermined timeframe, wherein an agent exhibits near-curb driving behavior when the agent operates within a particular distance of an edge of a road in the environment; and using the prediction to generate a planned path for a vehicle in the environment.