Trajectory Imputation via Temporal Decay and Mask Encoding
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for trajectory prediction in multi-agent scenarios assume complete observations, which is often not satisfied in real-world applications, leading to inaccurate predictions due to missing data.
Innovation Solution
A system that generates spatial and temporal missing patterns using a binary mask and Multi-Space Graph Neural Network (MS-GNN) to encode agent-wise spatial features, followed by a Conditional VRNN to model temporal dependencies, allowing for imputation and prediction of future trajectories from incomplete observations.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of manufacture
If existing methods assume complete observations of agent trajectories, then the modeling task is simplified, but prediction accuracy deteriorates in real-world scenarios with missing data
Solution Approach 1:
The patent introduces a missing pattern encoder as an intermediary component that processes binary masks representing missing observations. This encoder transforms the missingness information into meaningful features that are fed into the trajectory prediction model, allowing the system to adapt to incomplete data without requiring complete observations, thus resolving the contradiction between modeling simplicity and prediction accuracy under missing data conditions
Solution Approach 2:
The patent modifies the input parameters of the trajectory prediction model by incorporating missingness patterns (binary masks) as additional features. This parameter change allows the model to distinguish between observed and missing data points, adjusting its predictions accordingly. The model learns to weight observed and imputed trajectories differently based on their reliability, thereby maintaining accuracy even when observations are incomplete
2Adaptability or versatility
If trajectory prediction models process incomplete observational data, then real-world applicability improves, but modeling complexity increases
Solution Approach 1:
The patent segments the trajectory prediction task into distinct components: a missing pattern encoder that processes binary masks, an imputation module that generates missing trajectories, and a prediction module that combines observed and imputed data. This segmentation allows each component to handle specific aspects of incomplete data processing independently, reducing overall modeling complexity while improving real-world applicability
Solution Approach 2:
The patent performs preliminary encoding of missingness patterns before the main prediction task. The binary masks representing missing observations are encoded into features in advance, and imputed trajectories are generated beforehand based on these encoded patterns. This preliminary action prepares the model to handle incomplete data more efficiently during the actual prediction phase, reducing computational complexity while maintaining versatility
Data Source
AI summary
Systems and methods for trajectory imputation and prediction are provided. In one embodiment, a method includes generating a spatial missing pattern in an imputation stream by applying a binary mask to an observational dataset over a number of past timesteps. The method includes extracting spatial features from the spatial missing pattern for the number of past time steps. The method includes encoding the spatial features of the observational dataset into imputation latent variables in a latent space based on the spatial missing pattern. The method includes generating a temporal missing pattern by modeling temporal dependency as temporal decay from the past time to the first time based on the latent space. The method includes determining imputation trajectories based on the imputation latent variables and the temporal missing pattern. The method includes predicting future trajectories for the number of agents for a number of future timesteps based temporal missing pattern.