Agent Trajectory Forecasting With Uncertainty Propagation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current deep neural network (DNN) systems for object detection and trajectory forecasting in autonomous vehicles face challenges in accurately predicting agent trajectories and uncertainties, leading to suboptimal vehicle path determination and operation in complex traffic scenes.
Innovation Solution
The implementation of a variational autoencoder (VAE) that inputs agent locations and uncertainties, using convolutional neural networks (CNNs) and long short-term memory (LSTM) networks, to predict agent trajectories and potential trajectories, along with the use of a Kalman filter to determine uncertainties, improves the accuracy of trajectory forecasting by propagating state uncertainty and incorporating epistemic uncertainty from perception and trajectory forecasting.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If current deep neural network systems are used for object detection and trajectory forecasting, then the system can operate with simpler architecture, but the accuracy of predicting agent trajectories and uncertainties deteriorates
Solution Approach 1:
The system segments the trajectory forecasting task into multiple components: a perception module for detecting agents and their states, a trajectory forecasting module for predicting future positions, and an uncertainty estimation module for quantifying prediction confidence. This segmentation allows each module to specialize and improve overall accuracy while maintaining manageable complexity through modular design.
Solution Approach 2:
The patent introduces an intermediary uncertainty estimation mechanism that bridges the perception module and trajectory forecasting module. This intermediary component processes prediction uncertainties from the perception stage and propagates them through the forecasting process, enabling more accurate trajectory predictions by incorporating epistemic uncertainty without requiring a complete redesign of the system architecture.
2Reliability
If current DNN systems are used for trajectory forecasting, then the system operates faster with simpler processing, but the accuracy of vehicle path determination deteriorates
Solution Approach 1:
The system performs preliminary uncertainty estimation during the perception stage, before trajectory forecasting begins. By pre-computing uncertainty metrics for detected agents and their states, the system prepares quality indicators in advance that guide the trajectory forecasting process, improving path determination accuracy without adding significant processing time during critical real-time operation.
Solution Approach 2:
The patent implements a feedback mechanism where uncertainty estimates from the trajectory forecasting module are fed back to the perception and planning modules. This feedback loop allows the system to adjust its predictions and vehicle path determination based on confidence levels, improving reliability by continuously refining decisions based on prediction quality information.
Data Source
AI summary
A plurality of agent locations can be determined at a plurality of time steps by inputting a plurality of images to a perception algorithm that inputs the plurality of images and outputs agent labels and the agent locations. A plurality of first uncertainties corresponding to the agent locations can be determined at the plurality of time steps by inputting the plurality of agent locations to a filter algorithm that inputs the agent locations and outputs the plurality of first uncertainties corresponding to the plurality of agent locations. A plurality of predicted agent trajectories and potential trajectories corresponding to the predicted agent trajectories can be determined by inputting the plurality of agent locations at the plurality of time steps and the first uncertainties corresponding to the agent locations at the plurality of time steps to a variational autoencoder. The plurality of predicted agent trajectories and the potential trajectories corresponding to the predicted agent trajectories can be output.