Multi-Way Stop Precedence Using Occluded Agent Re-Identification
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Autonomous vehicles face challenges in determining the precedence order at multi-way stops, particularly when agents are occluded or execute rolling stops, leading to potential conflicts and inefficiencies in navigation.
Innovation Solution
The system assigns unique identifications to observed agents and re-identifies previously occluded agents using sensor data from LiDAR, cameras, and radar, applying rules such as FIFO, yield-to-the-right, and Straight, Near, Far, U-turn to determine precedence order, even in cases of occlusions and rolling stops.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If the system uses basic sensor detection at multi-way stops, then the navigation process is simple, but the system cannot accurately determine precedence order when agents are occluded or execute rolling stops
Solution Approach 1:
The system performs preliminary actions by assigning unique identification tags to all detected agents before the stop event, and pre-determines their initial positions and trajectories. This preliminary data collection enables accurate precedence determination even when agents are later occluded or behave non-compliantly, without requiring complex real-time analysis during the critical decision moment
Solution Approach 2:
The patent introduces an intermediary computational layer that processes sensor data from multiple sources (LiDAR, cameras, radar) and synthesizes a unified representation of agents, their positions, and their intended trajectories. This intermediary processing layer acts as a mediator between raw sensor inputs and the final precedence determination, enabling accurate conflict resolution without directly increasing sensor complexity
2Reliability
If the system continuously tracks all agents at the stop, then the precedence order can be accurately determined, but the computational load and processing time increase
Solution Approach 1:
The system extracts and focuses computational resources only on agents that are relevant to the current stop event and potential conflict zones. By filtering out irrelevant agents and concentrating tracking efforts on those within the stop vicinity and their projected paths, the system maintains high reliability in precedence determination while minimizing unnecessary computational overhead and processing time
Solution Approach 2:
The system performs preliminary identification and tagging of agents before they reach the stop line, establishing their initial trajectories and potential conflict relationships in advance. This preliminary action allows the system to predict which agents will need detailed tracking and resource allocation, reducing real-time computational burden while maintaining accurate precedence determination
3Loss of information
If the system only detects visible agents, then the detection process is straightforward, but occluded agents cannot be identified leading to potential conflicts
Solution Approach 1:
The patent introduces sensor fusion as an intermediary mechanism that combines data from multiple sensor types (LiDAR for depth, cameras for visual identification, radar for penetration through obstacles) to detect and track agents even when partially occluded. This intermediary fusion process reconstructs information about occluded agents by correlating data from different modalities, reducing information loss without requiring each individual sensor to achieve impossible detection levels
Solution Approach 2:
The system performs preliminary detection and assignment of unique identification tags to agents before they become occluded by other vehicles or objects. By establishing the agent's identity, position, and trajectory in advance while still visible, the system maintains continuous tracking through occlusion periods using predictive algorithms, preventing information loss about occluded agents
4Adaptability or versatility
If the system applies multiple precedence rules (FIFO, yield-to-right, Straight Near Far U-turn), then the navigation compliance improves, but the decision-making complexity increases
Solution Approach 1:
The patent segments the complex precedence determination into distinct, modular rule modules (FIFO module, yield-to-right module, Straight-Near-Far-U-turn module). Each module independently evaluates its specific criterion and outputs a preliminary precedence assessment. The system then integrates these segmented evaluations in a hierarchical manner, where higher-priority rules override lower-priority ones, achieving comprehensive traffic rule compliance while keeping each individual rule module simple and maintainable
Data Source
AI summary
Among other things, techniques are described for determining precedence order at a multiway stop. In embodiments, identifications are assigned to tracks, and young tracks are compared to stale tracks. A young track matches a stale track based on one or more factors. An identification of the young track is reassigned to an identification of the stale track, wherein the young track is determined to match the stale track based on the one or more factors. An earliest time of appearance of agents is determined based on identifications and in view of perception obscured areas. A precedence order for navigating through the intersection is determined based on local rules, the identifications, and the earliest time of appearance of agents, and the vehicle proceeds through the multiway stop intersection in accordance with the precedence order.