Autonomous Vehicle Trajectory Planning via Factor Graph Segmentation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Autonomous vehicles face challenges in planning safe trajectories on highways due to dynamic environments and high speeds, requiring accurate estimation of other vehicles' positions and intentions while ensuring smooth control signals for efficiency.
Innovation Solution
A method involving onboard systems that estimate past states of observed vehicles using sensor inputs, predict future trajectories, and plan the host vehicle's trajectory based on safety and control cost functions, utilizing graphical model-based factorization to account for collision avoidance and smooth control signals.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If the host vehicle travels at high speed on highway, then productivity is improved, but the environment becomes highly dynamic making motion planning more difficult
Solution Approach 1:
The motion planning problem is segmented into multiple independent trajectory hypotheses (e.g., maintaining current lane, changing to left lane, changing to right lane). Each hypothesis is evaluated separately using factor graphs, allowing the complex high-speed highway planning problem to be broken down into manageable sub-problems that can be solved independently and then compared.
Solution Approach 2:
The system dynamically adapts to the high-speed highway environment by continuously updating trajectory predictions and re-evaluating multiple hypotheses in real-time. The factor graph framework allows dynamic incorporation of new sensor data and changing environmental conditions, enabling the planning system to remain effective despite the dynamic nature of high-speed travel.
2Reliability
If the system predicts future trajectories of multiple observed vehicles, then collision avoidance is improved, but computational complexity increases
Solution Approach 1:
The prediction problem is segmented by creating separate trajectory hypotheses for each observed vehicle. Instead of solving one complex multi-vehicle prediction problem, the system independently predicts trajectories for each vehicle under different intention scenarios (e.g., vehicle A may continue straight, change left, or change right), then combines these predictions in the factor graph evaluation.
Solution Approach 2:
The system changes parameters by discretizing continuous trajectory possibilities into a finite set of hypotheses. By representing vehicle intentions as discrete choices (maintain lane, change to specific lane) rather than continuous variables, the computational complexity is reduced while maintaining sufficient accuracy for collision avoidance.
3Reliability
If the motion planning considers multiple objectives including collision avoidance and smooth control, then reliability is improved, but the planning problem becomes more complex
Solution Approach 1:
Multiple objectives (collision avoidance, smooth control, trajectory feasibility) are merged into a single unified factor graph framework. Different cost functions for each objective are combined as separate factors in the graph, allowing the system to evaluate multiple criteria simultaneously through a single integrated optimization process rather than handling each objective separately.
Solution Approach 2:
The system transforms the multi-objective planning problem into a parameter optimization problem by assigning weights to different cost factors. By changing the problem formulation from handling multiple objectives as separate constraints to optimizing a weighted sum of cost functions, the complexity is reduced while maintaining the ability to satisfy multiple objectives.
Data Source
AI summary
Aspects of the disclosure are related to a method, apparatus, and system for planning a motion for a first vehicle, comprising: estimating past states of an observed second vehicle based on sensor inputs; predicting a future trajectory of the second vehicle based on the estimated past states; planning a future trajectory of the first vehicle based on the predicted future trajectory of the second vehicle and a safety cost function; and driving the first vehicle to follow the planned trajectory.