Autonomous Vehicle Trajectory Prediction via Raster Scoring
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional approaches to autonomous vehicle navigation face challenges in accurately predicting the trajectories of surrounding objects, leading to potential collisions and inefficiencies in decision-making.
Innovation Solution
A system that generates a raster representation of the vehicle's environment, using machine learning models to predict object trajectories by proposing multiple trajectory options, scoring their likelihood, and modifying these proposals with offsets to improve accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional trajectory prediction methods are used, then the system is simpler to implement, but the prediction accuracy is insufficient leading to potential collisions
Solution Approach 1:
The trajectory prediction is segmented into multiple discrete trajectory proposals (e.g., going straight, turning left, turning right) rather than attempting to predict a single continuous trajectory. Each proposal is independently evaluated by the machine learning model, allowing for more precise and accurate predictions while maintaining manageable system complexity through modular processing
Solution Approach 2:
The system dynamically selects the most likely trajectory from multiple proposals based on real-time scoring by the machine learning model. The trajectory proposals are generated dynamically based on current environmental context and object states, allowing the system to adapt to changing conditions and improve prediction accuracy without requiring a completely complex rigid system
2Measurement precision
If multiple trajectory proposals are generated and evaluated, then the prediction accuracy improves, but the computational time and processing complexity increase
Solution Approach 1:
The system generates a limited set of discrete trajectory proposals (typically 3-5 common trajectories) rather than evaluating all possible continuous trajectories. This partial action approach focuses computational resources on the most relevant and likely trajectories, achieving high prediction accuracy while avoiding excessive computational time requirements
Solution Approach 2:
The machine learning model uses parameter changes in the form of scoring different trajectory proposals based on learned patterns from training data. By transforming the prediction problem into a scoring task with adjustable parameters, the system can efficiently evaluate multiple trajectories and select the best one without requiring exhaustive computational analysis of each possibility
Data Source
AI summary
Systems, methods, and non-transitory computer-readable media can determine a raster representative of a surrounding environment of a vehicle, wherein the raster depicts one or more objects in the surrounding environment of the vehicle. A plurality of trajectory proposals are determined for a first object of the one or more objects. For each trajectory proposal of the plurality of trajectory proposals, a score indicative of a likelihood that the first object will take a trajectory consistent with the trajectory proposal, and an offset for modifying the trajectory proposal are generated. A predicted trajectory is determined for the first object based on the scores and the offsets for the plurality of trajectory proposals.