Autonomous Driving Obstacle Response Using Confidence and Distance
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Autonomous driving vehicles face challenges in determining the appropriate response to obstacles due to uncertainty in obstacle movement and sensing reliability, leading to potential safety issues and unnecessary delays or disturbances in driving experience.
Innovation Solution
A method for operating autonomous driving vehicles that calculates a confidence score for obstacles based on sensing data, delaying reactions if the confidence score is below a threshold and the distance to the obstacle is greater than a certain threshold, allowing for safer and more efficient navigation by reducing unnecessary interventions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If the ADV reacts immediately to sensed obstacles, then safety is improved, but unnecessary delays and disturbances in driving experience occur when confidence is low
Solution Approach 1:
The system changes the response parameter from binary (react/don't react) to a continuous decision based on confidence score thresholds. By introducing a configurable confidence threshold parameter, the system adapts its reaction behavior dynamically, reacting only when confidence exceeds the threshold, thus balancing safety with smooth driving experience.
Solution Approach 2:
The system makes the reaction decision dynamic by continuously evaluating the confidence score and comparing it against the threshold. The decision to react or delay is not static but adapts in real-time based on the changing confidence level from sensor data, allowing flexible response to varying obstacle situations.
2Loss of time
If the ADV delays reaction to obstacles, then unnecessary interventions are reduced, but safety may be compromised if the obstacle is real and close
Solution Approach 1:
The system uses the distance parameter in combination with confidence score to make the delay decision. By introducing distance as a critical parameter with its own threshold, the system ensures that even if confidence is high, reactions are delayed only when distance is sufficient, preventing unsafe delays in close-proximity situations.
Solution Approach 2:
The system dynamically adjusts the delay duration based on real-time distance measurements. As the obstacle approaches and distance decreases, the system's ability to delay diminishes, ensuring that safety takes precedence over smooth driving experience when the obstacle is close enough to pose an immediate threat.
3Reliability
If the ADV reacts to all sensed obstacles, then safety is maximized, but false positives and unnecessary reactions increase
Solution Approach 1:
The system uses confidence score as a feedback mechanism from the sensor data processing. By evaluating the confidence level that the sensed object is a real obstacle before triggering a reaction, the system filters out false positives while maintaining sensitivity to genuine threats, thus reducing unnecessary reactions without compromising safety.
4Object-generated harmful factors
If the ADV waits for higher confidence, then false positives are reduced, but response time to real obstacles increases
Solution Approach 1:
The system balances false positive reduction with response time by introducing two parameters: confidence threshold and distance threshold. The interaction of these parameters creates an optimized decision boundary that reduces false positives while ensuring that response time remains acceptable for genuine obstacles within safe distance.
Data Source
Figure 1
Figure 2
Figure 3A~3B
AI summary
In one embodiment, a process is performed during controlling Autonomous Driving Vehicle (ADV). A confidence level associated with a sensed obstacle is determined (404). If the confidence level is below a confidence threshold, and a distance between the ADV and a potential point of contact with the sensed obstacle is below a distance threshold, then performance of a driving decision is delayed (412). If the confidence level is not below the confidence threshold, or the distance between the ADV and the potential point of contact with the sensed obstacle is not below the distance threshold, the driving decision is performed (410) to reduce risk of contact with the sensed obstacle.