Vehicle Collision Avoidance Planning in Constricted Passing Areas
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current prediction techniques for autonomous vehicles fail to adapt to space constrictions in environments, leading to the inability of vehicles to pass each other without colliding.
Innovation Solution
A collision avoidance algorithm that utilizes geometric and machine learning models to predict collision probabilities and adjust vehicle trajectories by simulating 'stepping backward' and 'stepping forward' to avoid collisions, allowing the vehicle to slow down or stop without requiring lane changes.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If current prediction techniques are used, then computational simplicity is maintained, but the ability to adapt to space constrictions and avoid collisions is insufficient
Solution Approach 1:
The prediction technique is segmented into multiple independent trajectory hypotheses, each representing a possible future state of the other vehicle. This allows the system to evaluate multiple scenarios separately and combine results, improving collision detection reliability while maintaining computational manageability through modular processing
Solution Approach 2:
The system performs preliminary prediction of multiple trajectories before making final collision determination. By pre-calculating possible paths and their intersection probabilities, the system prepares collision avoidance decisions in advance, improving response reliability without requiring complex real-time computation during critical moments
2Reliability
If the vehicle slows down or stops to avoid collision, then safety is improved, but traffic flow efficiency deteriorates
Solution Approach 1:
The system dynamically adjusts vehicle trajectory based on real-time collision probability assessments. Rather than static slow-down protocols, the vehicle continuously adapts its path using predicted trajectories of other vehicles, improving safety outcomes while minimizing unnecessary stops that would reduce traffic flow efficiency
Solution Approach 2:
The system changes key trajectory parameters (position, velocity, acceleration) based on predicted intersections with other vehicles. By adjusting these parameters dynamically according to collision risk assessment, the system achieves safer outcomes while maintaining optimal traffic flow through precise, minimal interventions rather than blanket speed reductions
3Measurement precision
If geometric models and machine learning models are used together, then collision prediction accuracy is improved, but computational complexity increases
Solution Approach 1:
The computational system is segmented into distinct geometric modeling and machine learning components that operate independently. The geometric model handles trajectory intersection calculations while the machine learning model predicts other vehicle behavior patterns. This segmentation improves prediction accuracy through specialized processing while managing complexity by avoiding integrated monolithic architecture
Solution Approach 2:
The system uses predicted trajectories as an intermediary between geometric collision detection and machine learning behavior prediction. This intermediary layer allows both models to contribute to collision probability assessment without requiring direct complex interaction between them, improving overall prediction accuracy while maintaining manageable computational complexity through modular information flow
Data Source
Figure 1
Figure 2
Figure 3
AI summary
Techniques for controlling a vehicle based on a collision avoidance algorithm are discussed herein. The vehicle receives sensor data and can determine that the sensor data represents an object in an environment through which the vehicle is travelling. A computing device associated with the vehicle determines a collision probability between the vehicle and the object at predicted locations of the vehicle and object at a first time. Updated locations of the vehicle and object can be determined, and a second collision probability can be determined. The vehicle is controlled based at least in part on the collision probabilities.