Object Path Prediction Architecture for Diverse Urban Trajectories

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Autonomous vehicles face challenges in predicting the future paths of objects in dense urban environments due to high computational demands and the difficulty in identifying relevant objects, accurately predicting object reactions, and handling erratic behavior, which complicates safe and efficient operation.

Innovation Solution

A machine-learned model architecture predicts diverse time-invariant paths for objects by using gradient descent on the closest path to ground truth, incorporating top-down environmental representations, and classifying objects as active or inactive to optimize computational resources.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Reliability

If the autonomous vehicle tracks and predicts all detected objects in dense urban environments, then the completeness of object monitoring is improved, but the computational resource consumption increases significantly

Engineering Contradiction:

Improveobject monitoring completenessVSAvoidcomputational resource consumption

Core Design Contradiction:

ReliabilityVSUse of energy by moving object

Solution Approach 1:

The patent segments objects into two categories: active objects that require path prediction and inactive objects that do not. This segmentation allows the system to focus computational resources only on relevant objects, reducing overall computational consumption while maintaining monitoring completeness for all detected objects.

Inventive Principle:

Principle #1Segmentation

Solution Approach 2:

The patent applies partial action by performing path prediction only on a subset of objects (active objects) rather than all detected objects. The system selectively applies the computationally intensive path prediction algorithm only where necessary, based on object characteristics and relevance to the autonomous vehicle's operation.

Inventive Principle:

Principle #16Partial or excessive action

2Measurement precision

If the autonomous vehicle uses complex path prediction algorithms to predict all possible object behaviors including rare erratic movements, then the prediction accuracy for adversarial maneuvers is improved, but the computational latency increases

Engineering Contradiction:

Improveprediction accuracyVSAvoidcomputational latency

Core Design Contradiction:

Measurement precisionVSLoss of time

Solution Approach 1:

The patent segments path prediction into two stages: time-invariant path prediction for general object trajectories and time-variant trajectory prediction for refined positional accuracy. This segmentation allows the system to first determine the relevant path efficiently, then apply more computationally intensive temporal predictions only when needed, reducing overall latency while maintaining accuracy for adversarial maneuvers.

Inventive Principle:

Principle #1Segmentation

Solution Approach 2:

The patent performs time-invariant path prediction as a preliminary step before time-variant trajectory prediction. By first determining the spatial path an object will follow, the system can then focus computational resources on predicting timing and positional variations along that path, rather than simultaneously computing all aspects of prediction, thus reducing computational latency.

Inventive Principle:

Principle #10Preliminary action

3Adaptability or versatility

If the autonomous vehicle stores and processes diverse path predictions for multiple objects, then the ability to handle adversarial maneuvers is improved, but the storage requirements increase

Engineering Contradiction:

Improveadversarial maneuver handlingVSAvoidstorage requirements

Core Design Contradiction:

Adaptability or versatilityVSQuantity of substance

Solution Approach 1:

The patent segments path storage by maintaining diverse time-invariant paths only for active objects that are relevant to the autonomous vehicle's operation, rather than storing paths for all detected objects. This segmentation reduces storage requirements while maintaining the system's ability to handle adversarial maneuvers involving relevant objects.

Inventive Principle:

Principle #1Segmentation

4Measurement precision

If the autonomous vehicle predicts time-variant trajectories with high temporal resolution, then the precision of object position prediction over time is improved, but the computational complexity increases

Engineering Contradiction:

Improvetemporal position prediction precisionVSAvoidprediction system complexity

Core Design Contradiction:

Measurement precisionVSDevice complexity

Solution Approach 1:

The patent segments the prediction system into two independent components: time-invariant path prediction that determines the spatial trajectory, and time-variant trajectory prediction that refines temporal positioning along the path. This segmentation reduces overall system complexity by allowing each component to specialize in one aspect of prediction, while maintaining high temporal resolution precision through the coordinated operation of both components.

Inventive Principle:

Principle #1Segmentation

Data Source

PatentUS12559138B2Machine-learned model architecture for diverse object path prediction

Publication Date: 2026.02.24 ZOOX INC

AI summary

A machine-learned architecture may predict a set of spatially-diverse paths that an object may take in the future. The paths generated by this architecture may be time-invariant (e.g., not identifying a time at which the object may occupy a position along one of these paths) but can be used by a second machine-learned model to predict progress in time along these paths. This segregation of the spatial paths and progress in time along the paths improves the accuracy of the ultimate prediction and better captures rare object behavior.