Vehicle Trajectory Prediction Using Hybrid Horizon Extension

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Existing autonomous vehicle (AV) systems struggle with limited prediction horizons of about 4 to 8 seconds, which is insufficient for advanced motion planning requiring longer forecasts of object trajectories, and existing machine learning models require extensive computational resources and retraining for horizon extension.

Innovation Solution

A hybrid approach combining a deep learning model for a first prediction horizon with a physical model for a second consecutive horizon, using predicted states from the deep learning model as input to extend the overall prediction to 8-15 seconds, enabling efficient on-board real-time trajectory planning.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If a machine learning model is used for trajectory prediction, then prediction accuracy is improved, but the prediction horizon is limited to 4-8 seconds

Engineering Contradiction:

Improveprediction accuracyVSAvoidprediction horizon

Core Design Contradiction:

Measurement precisionVSDuration of action of moving object

Solution Approach 1:

The prediction horizon is segmented into two parts: a first prediction horizon (4-8 seconds) handled by a machine learning model for accurate short-term prediction, and a second prediction horizon (beyond 8 seconds) handled by a physics-based model for extended long-term prediction. This segmentation allows each model to operate within its optimal range, resolving the contradiction between accuracy and horizon length.

Inventive Principle:

Principle #1Segmentation

2Adaptability or versatility

If the prediction horizon is extended beyond 8 seconds, then motion planning capability is improved, but computational resources and retraining requirements increase

Engineering Contradiction:

Improvemotion planning capabilityVSAvoidcomputational resources

Core Design Contradiction:

Adaptability or versatilityVSDevice complexity

Solution Approach 1:

The system segments the prediction task between a machine learning model (trained once on historical data) and a physics-based model (requiring no training). The ML model handles the computationally intensive short-term prediction, while the physics model extends the horizon with minimal computational overhead, thus extending planning capability without proportionally increasing complexity.

Inventive Principle:

Principle #1Segmentation

Solution Approach 2:

The machine learning model acts as an intermediary that processes sensor data and initial trajectory predictions, then passes results to the physics-based model for extension. This intermediary approach allows the system to leverage the strengths of both models while distributing computational load appropriately.

Inventive Principle:

Principle #24Intermediary (Mediator)

Data Source

PatentUS12409864B2Methods and systems for long-term trajectory prediction by extending a prediction horizon

Publication Date: 2025.09.09 FORD GLOBAL TECH LLC

AI summary

Systems and methods for planning a trajectory for a vehicle are disclosed. The methods include generating first predicted states of an object in an environment of the vehicle at one or more first time steps during a first prediction horizon using a machine learning model and sensor data about the environment. The first predicted states are used as an input to a physical model to generate second predicted states of the object at one or more second time steps during a second prediction horizon where the one or more second time steps are after the one or more first time steps. A trajectory for the vehicle is generated using the first predicted states or the second predicted states of the object.