Pedestrian Motion Prediction Grids for Autonomous Vehicle Path Planning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Autonomous vehicles face challenges in predicting the motion of diverse and unpredictable pedestrians, such as those who are partially occluded, change direction quickly, and traverse various terrains, due to suboptimal data quality and complexity in trajectory modeling.
Innovation Solution
A grid-based prediction method is employed, where a grid is projected around detected pedestrians to predict their possible future locations within a short time frame, using observed speed, direction, and orientation, with high-probability cells forming a heat map that helps in path planning and contour generation to avoid potential areas.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If learned trajectory proposal based behavior models are used to predict pedestrian motion, then prediction accuracy can be improved when data quality is high, but the system becomes overly complex and computationally intensive when dealing with diverse and unpredictable pedestrian behaviors
Solution Approach 1:
The patent segments the continuous trajectory prediction problem into discrete grid cells around the pedestrian. Instead of modeling continuous trajectories, the system divides space into a grid and predicts the probability of the pedestrian entering each cell within a time window. This segmentation simplifies the computational complexity while maintaining prediction accuracy for diverse pedestrian behaviors.
Solution Approach 2:
The patent uses a computationally inexpensive grid-based probability map that is regenerated frequently (short-lived) rather than relying on complex, pre-trained trajectory models. Each grid cell stores a simple probability value that is quickly updated based on current pedestrian state, avoiding the need for extensive training data and complex model architectures.
2Measurement precision
If extensive training data is collected to improve pedestrian behavior prediction, then model accuracy improves, but data collection time and processing complexity increase significantly
Solution Approach 1:
The system uses the pedestrian's own observed motion data (speed, direction, orientation) to generate predictions in real-time, rather than relying on pre-collected training data from extensive datasets. The grid-based probability calculation is performed autonomously using current sensor data, eliminating the need for separate data collection and training phases.
Solution Approach 2:
The patent pre-defines the grid structure and probability calculation methodology before operation, so that during runtime, only simple probability computations are needed based on current pedestrian state. This preliminary setup avoids the need for time-consuming data collection and model training during deployment.
3Reliability
If the vehicle maintains a larger safety distance from pedestrians to avoid collisions, then collision risk decreases, but the vehicle's mobility and efficiency are reduced
Solution Approach 1:
The patent applies different safety considerations to different spatial locations by computing cell-by-cell probability maps around each pedestrian. The vehicle can maintain larger distances in high-probability cells (where collision risk is high) while allowing closer proximity in low-probability cells, optimizing the balance between safety and mobility dynamically based on local risk assessment.
Solution Approach 2:
The safety margin is not fixed but dynamically adjusted based on real-time probability predictions. The vehicle continuously updates the grid-based probability map and adjusts its path planning accordingly, allowing it to take calculated risks in low-risk areas while maintaining conservative distances in high-risk areas, thereby improving overall mobility without compromising safety.
Data Source
AI summary
The technology relates to controlling a vehicle in an autonomous driving mode. For instance, sensor data identifying an object in an environment of the vehicle may be received. A grid including a plurality of cells may be projected around the object. For each given one of the plurality of cells, a likelihood that the object will enter the given one within a period of time into the future is predicted. A contour is generated based on the predicted likelihoods. The vehicle is then controlled in the autonomous driving mode in order to avoid an area within the contour.