Vectorized Trajectory Prediction Without Rendered Scene Images

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Existing systems for predicting the future trajectory of agents in an environment, such as autonomous vehicles, face inefficiencies due to the use of rendered images, which require computationally intensive encoding and result in lossy rendering, and are not optimal for learning trajectory predictions.

Innovation Solution

The system employs vectorized representations of scenes, approximating geographic entities and agent dynamics using polylines represented as sequences of vectors, allowing for efficient generation of trajectory predictions by leveraging spatial locality and high-order interactions among scene components.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Loss of information

If rendered images are used to represent the scene, then the system can capture visual information, but the encoding becomes computationally intensive and results in lossy rendering

Engineering Contradiction:

Improverendering lossVSAvoidcomputational power

Core Design Contradiction:

Loss of informationVSPower

Solution Approach 1:

The patent extracts only the essential geometric and semantic information from the scene by representing geographic entities as polylines and agents as vector sequences, eliminating the need for full rendered image encoding while preserving critical trajectory prediction data

Inventive Principle:

Principle #2Taking out (Extraction)

Solution Approach 2:

The patent changes the representation parameters from pixel-based rendered images to vector-based polylines and sequences, transforming the data format to reduce computational complexity while maintaining predictive accuracy

Inventive Principle:

Principle #35Parameter changes

2Reliability

If rendered images are used as input, then the scene can be visually represented, but the model requires significantly more parameters and FLOPs

Engineering Contradiction:

Improvetrajectory prediction accuracyVSAvoidmodel parameters

Core Design Contradiction:

ReliabilityVSDevice complexity

Solution Approach 1:

The patent segments the scene into distinct geometric components (polylines for geographic entities, vector sequences for agents) that can be processed independently and efficiently, reducing the overall model complexity while maintaining prediction reliability

Inventive Principle:

Principle #1Segmentation

3Productivity

If rendered images are used, then the system can process scene data, but the processing becomes computationally intensive

Engineering Contradiction:

Improveprocessing efficiencyVSAvoidcomputational power

Core Design Contradiction:

ProductivityVSPower

Solution Approach 1:

The patent replaces the mechanical image rendering and encoding process with direct vector processing operations, substituting computationally intensive graphical processing with efficient mathematical operations on polyline and sequence data

Inventive Principle:

Principle #28Mechanics substitution (Replace mechanical system)

Data Source

PatentUS12217168B2Agent trajectory prediction using vectorized inputs

Publication Date: 2025.02.04 WAYMO LLC

AI summary

Methods, systems, and apparatus, including computer programs encoded on computer storage media, for agent trajectory prediction using vectorized inputs.