Autonomous Vehicle Trajectory Control Under Moving Object Uncertainty
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for predicting the trajectory of moving objects around autonomous vehicles are limited by the need for extensive data and computation resources, particularly in urban environments, and fail to integrate object intention and provide long-term predictions.
Innovation Solution
A method using object detection sensors to determine current dynamic information and covariance matrices, combined with environmental mapping data, to calculate candidate paths and occupancy of moving objects, allowing for real-time trajectory control without extensive training.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If neural networks and machine learning methods are used to predict trajectories, then predictions consistent with general driver behavior can be produced, but a large amount of input data is required which is a significant constraint in urban environments
Solution Approach 1:
The patent segments the trajectory prediction problem into multiple candidate trajectories rather than attempting to predict a single trajectory. This segmentation allows the system to consider multiple possible future paths simultaneously, each with its own probability, thereby improving prediction reliability without requiring exponentially more data to cover all possible scenarios.
Solution Approach 2:
The patent generates a limited set of candidate trajectories (e.g., 3-5 most likely paths) rather than exhaustively predicting all possible trajectories. This partial action approach provides sufficient prediction accuracy for safety-critical applications while significantly reducing computational resources and data requirements compared to comprehensive prediction methods.
2Productivity
If model-based prediction methods are used, then computation resources are reduced, but the prediction horizon is limited and cannot provide long-term predictions necessary for good anticipation
Solution Approach 1:
The patent divides the prediction horizon into multiple time steps or stages, where each stage predicts trajectories for a shorter duration using efficient model-based methods. By segmenting the long-term prediction into sequential short-term predictions, the system achieves extended prediction horizons while maintaining computational efficiency at each stage.
Solution Approach 2:
The patent performs preliminary classification of vehicle maneuvers (e.g., lane keeping, lane changing, turning) before detailed trajectory prediction. This preliminary action identifies the most likely maneuver category, which then constrains subsequent trajectory generation to relevant candidate paths, extending the effective prediction horizon without proportionally increasing computational resources.
3Duration of action of moving object
If maneuver classification is combined with model-based prediction, then the prediction horizon is increased, but possible maneuvers are limited to lane following, lane changing, or turning which may be insufficient in urban conditions
Solution Approach 1:
The patent implements a dynamic maneuver set that adapts to the current driving context and environment type. In urban environments, the system dynamically expands the set of possible maneuvers to include complex actions like U-turns, three-point turns, and intersections with pedestrians, whereas in highway environments it uses the simpler classification set. This dynamic adaptation maintains prediction horizon extension benefits while ensuring versatility for diverse maneuver types.
Solution Approach 2:
The patent changes the parameters defining available maneuvers based on environmental context, vehicle type, and road geometry. For example, the system adjusts the set of candidate maneuvers according to whether the vehicle is in an urban or rural environment, the type of road intersection, and the detected presence of vulnerable road users, thereby achieving both extended prediction horizon and comprehensive maneuver coverage through parameter adaptation.
Data Source
Figure 1
Figure 2
Figure 3~4
AI summary
The present invention relates to a method and a device for controlling the trajectory of an autonomous vehicle (10) travelling in an environment (1) comprising a moving object (11). For this purpose, dynamic information on the moving object (11) and an associated covariance matrix are obtained. A set of candidate paths (111, 112) of the moving object (11) are determined from the dynamic information. The moving object (11) is represented by a polygon considering the position and orientation uncertainties obtained from the covariance matrix. A maximum speed profile and a minimum speed profile are determined in order to determine the spatio-temporal occupation of the moving object (11) along each candidate path (111, 112). The trajectory of the autonomous vehicle (10) is controlled depending on the spatio-temporal occupation of the moving object (11) along each candidate path (111, 112).