Occupancy Flow Fields for Agent Identity and Velocity Prediction

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing occupancy grid-based methods for motion forecasting in autonomous vehicles fail to recover agent identity and cannot predict individual agent velocities directly, making it difficult to interpolate between time intervals and incorporate the behavior of undetected agents.

Innovation Solution

The integration of occupancy flow fields, which predict motion flow in addition to occupancy likelihood, allowing tracing of agent identity and enabling interpolation between time points using flow predictions.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Adaptability or versatility

If occupancy grid-based methods are used for motion forecasting, then occupancy predictions capturing rich future distributions are achieved, but agent identity recovery and velocity prediction become impossible

Engineering Contradiction:
Improveoccupancy prediction capabilityVSAvoidagent identity information
Core Design Contradiction:
Adaptability or versatilityVSLoss of information

Solution Approach 1:

The patent combines occupancy grid predictions with flow field predictions into a unified representation. The flow field adds vector information (dx, dy) to each grid cell, merging the scalar occupancy probability with directional motion information. This allows the system to maintain the rich distribution modeling of occupancy grids while recovering agent identity and velocity through the flow vectors that trace agent trajectories across time steps.

Inventive Principle:
Principle #5Merging (Combining)

2Device complexity

If occupancy grid predictions are generated without flow information, then computational simplicity is maintained, but interpolation between time intervals cannot be performed

Engineering Contradiction:
Improveprediction model complexityVSAvoidtime interpolation capability
Core Design Contradiction:
Device complexityVSProductivity

Solution Approach 1:

The flow field acts as an intermediary that bridges occupancy predictions at different time steps. By providing continuous motion vectors (dx, dy) for each grid cell, the flow field enables interpolation between discrete time intervals. The flow information serves as a mediator that allows the system to compute intermediate states by integrating velocity information, rather than relying solely on discrete occupancy snapshots.

Inventive Principle:
Principle #24Intermediary (Mediator)

3Reliability

If standard occupancy grids are used, then agent presence probability is modeled, but motion trajectory reconstruction is impossible

Engineering Contradiction:
Improveagent presence predictionVSAvoidmotion trajectory information
Core Design Contradiction:
ReliabilityVSLoss of information

Solution Approach 1:

The patent transitions from scalar occupancy values to vector-val flow fields by adding a new dimension of information. Instead of only predicting whether a cell is occupied, the system predicts occupancy probability combined with motion vectors (dx, dy) for each cell. This dimensional expansion from scalar to vector fields enables trajectory reconstruction while maintaining the reliable agent presence prediction capability of standard occupancy grids.

Inventive Principle:
Principle #17Another dimension (Dimensionality change)

Data Source

PatentEP4060613B1Predicting the future movement of agents in an environment using occupancy flow fields
Publication Date: 2025.07.02 WAYMO LLC
  • EP4060613B1 patent drawingFigure 1
  • EP4060613B1 patent drawingFigure 2
  • EP4060613B1 patent drawingFigure 3

AI summary

Methods, systems, and apparatus, including computer programs encoded on computer storage media, for predicting the future movement of agents in an environment. In particular, the future movement is predicted through occupancy flow fields that specify, for each future time point in a sequence of future time points and for each agent type in a set of one or more agent types: an occupancy prediction for the future time step that specifies, for each grid cell, an occupancy likelihood that any agent of the agent type will occupy the grid cell at the future time point, and a motion flow prediction that specifies, for each grid cell, a motion vector that represents predicted motion of agents of the agent type within the grid cell at the future time point.