Traffic Behavior Prediction Using Bayesian Trajectory Feedback
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current driving systems struggle to accurately predict traffic participant behavior beyond 2 seconds into the future, as they rely on simple kinematic extrapolations and do not effectively integrate past predictions with new observations, leading to prediction errors and computationally costly methods that often generate dynamically unfeasible trajectories.
Innovation Solution
A method that obtains real-time kinematic state distributions for traffic participants, projects these distributions into the future, and updates trajectory probabilities based on compatibility with observed kinematic states, using a Bayesian approach to efficiently predict motion and maneuver probabilities, accounting for interactions with the environment and other participants.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of time
If simple kinematic extrapolations are used for prediction, then the method is computationally simple, but the prediction is only valid up to approximately 2 seconds into the future
Solution Approach 1:
The system integrates tracking algorithms that continuously update predictions based on new observations. Past predictions are evaluated in light of whether they are supported by observations gathered since the prediction was made, creating a feedback loop that extends prediction validity beyond simple extrapolation limits
Solution Approach 2:
The system generates multiple candidate trajectories in advance and evaluates them against observations. By preparing multiple hypothesis trajectories beforehand and selectively validating them, the system can make longer-term predictions while maintaining accuracy through selective refinement of promising candidates
2Duration of action of moving object
If trajectory predictions are generated forward in time from current observations, then predictions can be made for longer time spans, but this leads to accumulation of prediction errors and computationally costly operations
Solution Approach 1:
The prediction process is divided into discrete time steps with intermediate evaluation points. Instead of generating one long-term prediction, the system segments the time horizon into multiple shorter intervals, evaluating trajectories at intermediate points to filter out erroneous predictions before they accumulate
Solution Approach 2:
The system generates more trajectory candidates than ultimately needed, then filters them through observation-based validation. By producing excessive candidate trajectories and selectively retaining only those consistent with observations, the system manages computational complexity through selective refinement rather than exhaustive processing
3Duration of action of moving object
If trajectory predictions are generated forward in time, then long-term predictions are possible, but this often leads to dynamically unfeasible trajectories
Solution Approach 1:
Observations of actual traffic participant behavior provide feedback that validates or rejects predicted trajectories. By continuously comparing predictions with actual observations, the system identifies and eliminates dynamically unfeasible trajectories from the candidate set
Solution Approach 2:
The system maintains multiple dynamic trajectory hypotheses that can be selectively activated based on observed behavior. Rather than committing to a single static prediction, the system dynamically adjusts which trajectory candidates remain viable based on ongoing observations, allowing adaptation to changing conditions while maintaining feasibility
Data Source
Figure 1
Figure 2
Figure 3
AI summary
The invention relates to a method (1) for predicting traffic participant behavior. The method (1) comprises obtaining a first kinematic state distribution of at least one traffic participant at a first time. The method (1) further comprises projecting second kinematic state distributions (13) of the at least one traffic participant at a second time, the second time being a first time span into the future from the first time. The method (1) also comprises defining a distribution of trajectories (14), wherein each trajectory of the distribution of trajectories (14) links a kinematic state of the first kinematic state distribution to a kinematic state of the second kinematic state distribution (13). Then, a third kinematic state distribution (15) of the at least one traffic participant at a third time is obtained, the third time being a second time span later than the first time, wherein the second time span is shorter than the first time span. Compatibilities between the third kinematic state distribution (15) and a distribution of kinematic states resulting from evaluating each of the trajectories of the distribution of trajectories (14) at the third time are determined and probabilities are assigned to the trajectories of the distribution of trajectories (14) based on the determined compatibilities. The invention further relates to a driving system and to a vehicle comprising a driving system.