Object Maneuver Prediction Using Road-Mapped Trajectory Scoring
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing self-driving car systems face challenges in performing safe maneuvers while avoiding collisions with dynamic objects in their vicinity, as they struggle to accurately predict the trajectories of these objects.
Innovation Solution
The system uses sensor data to generate and analyze predicted trajectories of dynamic objects without considering road lane boundaries, then maps these trajectories onto the road map to determine potential future location points, which are scored for association with road lanes, and an aggregated score is used to predict the object's location.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If the system generates predicted trajectories without considering road lane boundaries, then the system's ability to anticipate movements of dynamic objects is improved, but the accuracy of predicting object location within specific road lanes deteriorates
Solution Approach 1:
The system segments the trajectory prediction process into two independent stages: first generating unconstrained predicted trajectories that capture the full range of possible object movements, then separately mapping these trajectories onto road lane boundaries to determine lane-specific locations. This segmentation allows each stage to optimize for its specific purpose without compromise.
Solution Approach 2:
The system introduces an intermediary mapping process that bridges the gap between unconstrained predicted trajectories and road lane boundaries. This intermediary step projects trajectory points onto the road network, determining which road lane each point corresponds to, thereby reconciling the freedom of unconstrained prediction with the constraints of actual road geometry.
2Measurement precision
If the system maps predicted trajectories onto road map to determine potential future location points, then the accuracy of location prediction is improved, but the complexity of the processing system increases
Solution Approach 1:
The system performs preliminary action by generating the complete set of predicted trajectories first, without considering road constraints. This preliminary trajectory generation captures all possible object movements, which are then subsequently mapped onto the road map. This approach avoids the complexity of incorporating road constraints during the computationally intensive trajectory generation phase.
Solution Approach 2:
The processing system is segmented into distinct functional modules: a trajectory prediction module that generates unconstrained trajectories, and a mapping module that projects these trajectories onto road lanes. This modular segmentation reduces overall system complexity by allowing each module to be optimized independently for its specific task.
3Measurement precision
If the system computes scores for each potential future location point to determine association with road lanes, then the precision of lane association is improved, but the computational time and resources increase
Solution Approach 1:
The system applies partial action by computing scores only for the discrete set of potential future location points derived from the predicted trajectories, rather than continuously across the entire road lane. This selective scoring approach maintains lane association precision while significantly reducing computational burden compared to exhaustive methods.
Solution Approach 2:
The system performs preliminary action by first generating a finite set of discrete location points from the predicted trajectories before scoring them for lane association. This preliminary discretization reduces the computational space that needs to be evaluated, thereby reducing processing time while preserving association precision through the subsequent scoring of these pre-selected points.
Data Source
AI summary
Methods and devices for generating data for controlling a Self-Driving Car (SDC) are disclosed. The method includes: receiving a section of a road map corresponding to surroundings of the SDC and at least one object, generating predicted trajectories including potential future location points of the at least one object, mapping the potential future location points on the section of the road map, computing a score for each of the potential future location points, computing an aggregated score from the scores corresponding to the potential future location points, and based on the aggregated score, determining a predicted location of the at least one object at the future instance of time.