Vehicle Grid-Map Prediction for Multimodal Road User Trajectories
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for predicting future locations of objects in autonomous driving systems face challenges in efficiently and reliably handling multiple objects and complex driving scenarios, particularly due to imbalanced trajectory data and the need to account for multiple possible future positions and interactions among various types of road users.
Innovation Solution
A computer-implemented method using a grid map representation of road users' perception data and static environment data, combined with machine learning techniques such as neural networks, to predict properties like positions, speeds, and collision risk values, incorporating dynamic and static context encoders and decoders, and employing regression and classification methods to handle multiple object classes and future positions with varying probabilities.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional methods are used for predicting future locations of objects, then the system can handle standard trajectories, but it fails to accurately predict nonstandard trajectories and multimodal future positions in complex driving scenarios
Solution Approach 1:
The prediction task is segmented into multiple independent grid cells across the environment. Each grid cell independently predicts the probability of object presence at future time steps, allowing the system to handle complex scenarios with varying numbers and types of road users by processing local predictions that collectively form the global prediction map.
Solution Approach 2:
The patent introduces a temporal dimension to the prediction by generating predictions for multiple future time steps (t+1, t+2, etc.) simultaneously across the grid map. This allows the system to capture multimodal future positions and nonstandard trajectories by predicting object presence at different future times rather than a single fixed trajectory.
2Measurement precision
If the system predicts multiple possible future positions and trajectories, then prediction accuracy improves, but system complexity increases
Solution Approach 1:
Multiple predictions for different future time steps and object classes are merged into a single unified grid map structure. The grid map integrates predictions for various object types (vehicles, pedestrians, cyclists) and multiple future moments into one coherent representation, reducing system complexity by using a single data structure rather than separate prediction systems.
Solution Approach 2:
The grid map serves multiple functions simultaneously: it represents the environment, stores predictions for multiple future time steps, accommodates different object classes, and provides probability distributions for multimodal trajectories. This multi-functionality reduces the need for separate specialized systems for each prediction task.
3Adaptability or versatility
If the system handles varying numbers and types of road users, then adaptability improves, but computational requirements and system complexity increase
Solution Approach 1:
Each grid cell in the prediction map has local quality tailored to its specific location and context. The system predicts object presence independently for each grid cell based on local features and context, allowing it to adapt to varying numbers and types of road users without requiring a completely different prediction system for each scenario.
Solution Approach 2:
The system uses a template-based approach where the same grid map structure is copied and applied to different driving scenarios with varying road users. Rather than creating custom prediction systems for each type of road user or scenario, the universal grid map template can be instantiated for any combination of objects and environments.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
A computer implemented method for predicting properties of a plurality of objects in a vicinity of a vehicle comprises the following steps carried out by computer hardware components: determining a grid map representation of road users perception data, the road users perception data comprising tracked perception results and/ or untracked sensor intermediate detections; determining a grid map representation of static environment data based on data obtained from a perception system and/or a pre-determined map; and determining the properties of the plurality of objects based on the grid map representation of road users perception data and the grid map representation of static environment data.