Autonomous Driving Plans Using Two-Stage Road User Prediction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing autonomous vehicle systems face challenges in accurately predicting the actions of road users, leading to conservative driving behaviors that reduce efficiency and convenience, such as excessive slowing or stopping, due to incomplete road user prediction technology.
Innovation Solution
A method and device that utilize a combination of algorithm-based and data learning-based prediction models to generate accurate road user predictions, minimizing computation by focusing on significant road users, thereby improving driving plan determination for autonomous vehicles.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If road user prediction technology is not fully developed, then autonomous vehicles can travel more safely by conservatively predicting future states, but this leads to excessive control operations such as drastic velocity reduction, driving at excessively low velocity, or constant stopping
Solution Approach 1:
The prediction task is segmented into two stages: first, an algorithm-based model generates preliminary prediction data for all road users; second, a data learning-based model refines predictions for significant road users only. This segmentation allows the system to maintain safety through comprehensive initial assessment while improving efficiency by focusing computational resources on critical predictions, thereby reducing excessive conservative control operations.
Solution Approach 2:
The system changes the parameter of prediction accuracy by combining two different prediction models with different characteristics. The algorithm-based model provides a baseline prediction, while the data learning-based model adjusts and refines these predictions based on learned patterns from historical data. This parameter change enables more accurate prediction of road user future states, reducing the need for overly conservative safety margins that cause excessive velocity reduction and stopping.
2Measurement precision
If predictions are made about all road users through comprehensive models, then prediction accuracy improves, but the amount of computation increases significantly
Solution Approach 1:
The computation process is divided into two phases: a lightweight algorithm-based prediction for all road users that provides quick preliminary results, followed by a more intensive data learning-based prediction applied only to significant road users identified in the first phase. This segmentation maintains high prediction accuracy for critical cases while dramatically reducing overall computational load compared to applying comprehensive models to all road users.
Solution Approach 2:
Instead of applying the computationally intensive data learning-based model to all road users, the system performs partial action by focusing this intensive computation only on significant road users whose future states most impact autonomous vehicle safety and planning. This partial application of the more accurate model achieves sufficient prediction accuracy while minimizing computation amount.
Data Source
AI summary
Provided are a method determining a driving plan for an autonomous vehicle on the basis of road user prediction, a computing device for performing the same, and a recording medium on which a program for performing the same is recorded. The method performed by a computing device includes generating first prediction data including prediction results about a plurality of road users near an autonomous vehicle, generating second prediction data including only a prediction result about at least one of the plurality of road users, and determining a driving plan for the autonomous vehicle using the generated first prediction data and the generated second prediction data.