Pedestrian Trajectory Prediction Using Polar Occupancy Grids
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional pedestrian trajectory prediction systems for autonomous vehicles lack both accuracy and interpretability, with black box models providing high accuracy but low interpretability, and white box models being less accurate and more understandable.
Innovation Solution
A probabilistic motion model (PMM) using a polar occupancy grid to predict pedestrian trajectories, incorporating heatmap prior information and contextual factors to enhance prediction accuracy while maintaining interpretability, employing Bayesian filtering and kinematic assumptions to model pedestrian motion.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If black box models (deep learning, neural networks) are used for pedestrian trajectory prediction, then prediction accuracy is improved, but interpretability deteriorates
Solution Approach 1:
The prediction model is segmented into two distinct components: a black box model (neural network) for learning complex trajectory patterns from data, and a white box model (probabilistic motion model with dynamic polar occupancy grid) for providing interpretable predictions. Each component handles different aspects of the prediction task, allowing the system to leverage both high accuracy and interpretability without requiring a single monolithic model to do both.
2Device complexity
If white box models (dynamic and kinematic models) are used for pedestrian trajectory prediction, then interpretability is improved, but prediction accuracy deteriorates
Solution Approach 1:
The system merges a black box neural network model with a white box probabilistic motion model into a unified prediction framework. The neural network processes input features and generates predictions, while the probabilistic motion model with dynamic polar occupancy grid provides interpretable representations of pedestrian motion probabilities. Both models work together to produce predictions that are simultaneously accurate and interpretable.
Data Source
AI summary
A system for collision avoidance by an autonomous vehicle (AV) is disclosed. The system includes a sensor system to detect objects in the AV's proximity. The system also includes a pedestrian trajectory prediction (PTP) system with an input module to receive sensor data, including heatmap prior information about past pedestrian trajectories. A memory module stores the data, and a probabilistic motion model (PMM) module uses a dynamic polar occupancy grid to represent the pedestrian's predicted trajectory. The PMM calculates occupancy probabilities based on angular motion (related to heading) and radial motion (related to speed). A processing module combines these probabilities to predict the pedestrian's future position. If an imminent collision is detected, the system generates a warning signal output to the AV's user interface. This system enhances collision avoidance by combining past data and real-time sensor data with probabilistic trajectory predictions.