Autonomous Vehicle Drivability Classification for Path Planning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Autonomous vehicles face challenges in differentiating between objects that can be driven over and those that cannot, leading to unnecessary maneuvers such as abrupt stops or lane changes, which can affect safety and smooth operation.
Innovation Solution
A perception system that classifies objects as drivable or not drivable based on their characteristics, using sensor information and pre-stored map data, allowing the vehicle to determine whether it can safely drive over an object without causing damage, and adjust its path accordingly.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If the autonomous vehicle treats all detected objects as obstacles to be avoided, then safety is improved by preventing potential damage, but unnecessary maneuvers such as abrupt stops or lane changes occur reducing operational smoothness
Solution Approach 1:
The system applies different treatment rules to different types of objects based on their local characteristics. Drivable objects (debris, leaves, paper) are classified differently from non-drivable objects (animals, children, broken glass), allowing the vehicle to maintain smooth operation for safe objects while avoiding dangerous ones, thus resolving the contradiction between safety and operational smoothness
2Measurement precision
If the vehicle performs detailed classification of all detected objects, then accurate identification of drivable vs non-drivable objects is improved, but computational complexity and processing time increase
Solution Approach 1:
The object classification process is segmented into distinct stages: initial detection by perception system, filtering to identify objects in the vehicle's path, classification into drivable/non-drivable categories using processor analysis of sensor data, and subsequent maneuvering decisions. This segmentation allows complex classification to be performed only on relevant objects rather than all detected objects, reducing computational complexity while maintaining accuracy
3Reliability
If the vehicle alters its path to avoid all classified non-drivable objects, then safety is improved by preventing damage, but operational efficiency decreases due to frequent path alterations
Solution Approach 1:
The system changes the parameter of path alteration based on the classification result. For drivable objects, the path parameter remains unchanged (no avoidance maneuver). For non-drivable objects, the path parameter is modified to alter the trajectory and avoid the object. This conditional parameter change ensures safety-critical objects are avoided while maintaining operational efficiency for safe objects
Data Source
Figure 1
Figure 2
Figure 3A
AI summary
Aspects of the disclosure relate to maneuvering a vehicle. As an example, sensor information identifying a set of objects as well as a set of characteristics for each object of the set of objects is received from a perception system of a vehicle 100. The set of objects is filtered to remove objects corresponding to vehicles, bicycles, and pedestrians. An object within an expected future path of the vehicle is selected from the filtered set of objects. The object is classified as drivable or not drivable based on the set of characteristics. Drivable indicates that the vehicle can drive over the object without causing damage to the vehicle. The vehicle is maneuvered based on the classification such that when the object is classified as drivable, maneuvering the vehicle includes driving the vehicle over the object by not altering the expected future path of the vehicle.