Autonomous Vehicle Control Using Multi-Layer Actor Path Prediction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Autonomous vehicles face challenges in reliably interpreting and responding to their surroundings, particularly in complex environments, as existing technologies struggle to effectively process and utilize data from multiple actors and streams to make accurate navigation decisions.
Innovation Solution
A method for training machine learning models that processes data from actors and streams in an autonomous vehicle's environment, using multiple layers to generate predicted outputs and update model layers based on reference labels, enabling the vehicle to make informed navigation decisions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If multiple layers of machine learning models are used to process data from actors and streams, then the autonomous vehicle's decision-making capability and navigation accuracy are improved, but the computational complexity and processing time increase
Solution Approach 1:
The machine learning model is divided into multiple specialized layers, each processing specific aspects of environmental data (actors, streams, features). This segmentation allows complex decision-making to be broken down into manageable processing stages, improving reliability while organizing computational complexity into structured modules.
Solution Approach 2:
The patent processes data across multiple dimensional layers - spatial dimensions (actor positions, stream geometries), temporal dimensions (historical data, predictions), and feature dimensions (velocity, distance, pose). This multi-dimensional processing enhances decision-making capability by considering problems from multiple angles simultaneously.
2Measurement precision
If multiple layers of machine learning models are used to process data from actors and streams, then the autonomous vehicle's navigation accuracy is improved, but the data processing time increases
Solution Approach 1:
The system performs preliminary processing of actor and stream data before main decision-making, pre-computing features such as distances, velocities, and geometric relationships. This preliminary action reduces the computational burden during critical navigation decisions, improving accuracy while minimizing real-time processing time.
Solution Approach 2:
The patent replaces traditional rule-based navigation systems with machine learning models that learn optimal navigation strategies from data. This substitution enables more accurate navigation decisions by leveraging patterns recognized through training, achieving superior accuracy without relying on complex mechanical decision rules.
3Adaptability or versatility
If the machine learning model processes data from multiple actors and streams, then the autonomous vehicle's ability to handle complex environments is improved, but the system complexity increases
Solution Approach 1:
The machine learning model is designed as a universal system that processes diverse inputs (multiple actors, various stream types, different features) through a unified architecture. This multi-functional design enables the vehicle to handle various complex environmental scenarios using the same core system, improving adaptability while avoiding the need for separate specialized systems for each scenario.
Solution Approach 2:
The system dynamically adapts its processing based on environmental complexity, adjusting which actors and streams are processed in detail versus summarized. This dynamic behavior allows the system to handle complex environments effectively while managing computational resources efficiently, balancing versatility with system complexity.
Data Source
AI summary
Systems and methods related to controlling an autonomous vehicle (“AV”) are described herein. Implementations can process actor(s) from a past episode of locomotion of a vehicle, and stream(s) in an environment of the vehicle during the past episode to generate predicted output(s). The actor(s) may each be associated with a corresponding object in the environment of the vehicle, and the stream(s) may each represent candidate navigation paths in the environment of the vehicle. Further, implementations can process the predicted output(s) to generate further predicted output(s), and can compare the predicted output(s) to associated reference label(s). The processing can be performed utilizing layer(s) or distinct, additional layer(s) of machine learning (“ML”) model(s). Implementations can update the layer(s) or the additional layer(s) based on the comparing, and subsequently use the ML model(s) in controlling the AV.