Trajectory Forecasting with Dual-Horizon Neural Networks
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current learning methods for autonomous vehicle trajectory forecasting are limited by the need for good initialization parameters and fail to account for non-identical and non-independent data, particularly in structured physical environments with multiple interacting agents, leading to compounding errors.
Innovation Solution
A system that uses a neural network trained over multiple prediction horizons to encode both inertial and behavioral constraints, predicting object trajectories in a curvilinear coordinate system to improve motion planning for autonomous vehicles, incorporating perception data such as velocity, acceleration, and location information.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If a single prediction horizon is used for trajectory forecasting, then the model structure is simple, but it cannot simultaneously encode both inertial constraints (short-term) and behavioral constraints (long-term)
Solution Approach 1:
The prediction horizon is segmented into two distinct components: a first prediction horizon for encoding inertial constraints and a second prediction horizon for encoding behavioral constraints. This segmentation allows the model to separately process short-term physical dynamics and long-term behavioral patterns, resolving the contradiction between model simplicity and prediction accuracy by structuring complexity in a functionally meaningful way.
2Adaptability or versatility
If conventional learning methods are used for trajectory forecasting, then initialization with good starting parameters is required, but this increases the difficulty of model setup and reduces adaptability
Solution Approach 1:
The model performs preliminary action by pre-processing perception data into a curvilinear coordinate system before feeding it to the prediction model. This preliminary transformation aligns the data with the natural motion patterns of vehicles, allowing the model to learn more effectively without requiring careful initialization of parameters, thereby improving adaptability while reducing setup complexity.
3Measurement precision
If perception data is not transformed into curvilinear coordinate system, then processing is simpler, but prediction accuracy for vehicle trajectories is reduced
Solution Approach 1:
The perception data undergoes a parameter transformation from a standard Cartesian coordinate system to a curvilinear coordinate system that better represents vehicle motion dynamics. This change in coordinate parameters captures the inherent curvature and directional changes in vehicle trajectories, improving prediction accuracy while the transformation is designed to be computationally efficient.
Data Source
AI summary
Systems and methods for forecasting trajectories of objects. The method includes obtaining a prediction model trained to predict future trajectories of objects. The prediction model is trained over a first prediction horizon selected to encode inertial constraints in a predicted trajectory and over a second prediction horizon selected to encode behavioral constraints in the predicted trajectory. The method also include generating a planned trajectory of an autonomous vehicle by receiving state data corresponding to the autonomous vehicle, receiving perception data corresponding to an object, predicting a future trajectory of the object based on the perception data and the prediction model, and generating the planned trajectory of the autonomous vehicle based on the future trajectory of the object and the state data.