Latent Semantic Sampling for Diverse Vehicle Trajectory Prediction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing vehicle trajectory prediction systems lack diversity in their predictions, making it difficult to capture low-probability but high-importance outcomes, which is essential for safety verification and motion planning, especially at intersections where different maneuvers can have varying outcomes.
Innovation Solution
The method employs a latent semantic space using artificial neural networks to generate diverse vehicle trajectories by separating high-level and low-level representations, encouraging them to be uncorrelated while correlating the high-level representation with annotation coding, and using farthest point sampling to select a subset of latent state vectors for predicting vehicle trajectories.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If brute force repetition is used to improve diversity of trajectory predictions, then coverage of possible vehicle maneuvers is improved, but computational efficiency deteriorates and safety verification becomes difficult
Solution Approach 1:
The patent introduces an intermediary semantic space that mediates between the input trajectory data and the output diverse trajectories. By mapping trajectories to semantic categories (e.g., turning left, going straight, turning right) and then sampling from these semantic categories, the system achieves diverse trajectory generation without brute force repetition, resolving the contradiction between coverage and computational efficiency
Solution Approach 2:
The patent changes the parameter space from continuous trajectory coordinates to discrete semantic categories. By representing trajectories in terms of semantic parameters (maneuver types, road segments, intersections) rather than continuous spatial parameters, the system enables efficient sampling while maintaining comprehensive coverage of possible maneuvers
2Adaptability or versatility
If sampling is used to produce diverse trajectory outcomes, then coverage of possible maneuvers is improved, but detecting and avoiding collisions becomes computationally expensive
Solution Approach 1:
The patent segments the continuous space of possible trajectories into discrete semantic categories representing different maneuver types. By performing collision detection and safety verification at the semantic category level rather than for every possible continuous trajectory, the computational cost is dramatically reduced while still ensuring comprehensive safety coverage
3Measurement precision
If only the most likely trajectory is predicted, then prediction accuracy for the primary outcome is improved, but coverage of low-probability high-importance outcomes deteriorates
Solution Approach 1:
The patent implements a dynamic prediction system that adapts its output based on the prediction task requirements. For primary prediction, it outputs the most likely trajectory with high accuracy. For safety verification and planning, it dynamically switches to sampling multiple semantic categories to capture low-probability but high-importance outcomes, thus resolving the contradiction between precision and coverage
Data Source
AI summary
A system and method for generating a predicted vehicle trajectory includes a generative adversarial network configured to receive a trajectory vector of a target vehicle and generate a set of latent state vectors using the received trajectory vector and an artificial neural network. The latent state vectors each comprise a high-level sub-vector, ZH. The GAN enforces ZH to be correlated to an annotation coding representing semantic categories of vehicle trajectories. The GAN selects a subset, from the set of latent state vectors, using farthest point sampling and generates a predicted vehicle trajectory based on the selected subset of latent state vectors.