Autonomous Vehicle Object Prediction Using Actionable Uncertainty
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Autonomous vehicles face challenges in accurately navigating through environments with uncertain object detections and classifications, leading to potential hard braking or turning due to reliance on single best detections without considering uncertainty.
Innovation Solution
The system generates and shares uncertainty data between components of an autonomous vehicle's computing system, using machine-learned models to detect objects, track their states, and classify them, while also determining uncertainty associated with these detections and predictions, which are then used to improve object prediction and motion planning.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If the autonomous vehicle relies on single best detections without considering uncertainty, then the navigation response is fast and decisive, but the vehicle experiences hard braking or turning due to inaccurate detections
Solution Approach 1:
The patent introduces an uncertainty quantification module as an intermediary between the object detection system and the motion planning system. This module processes detection results and generates uncertainty metrics that mediate the decision-making process, allowing the system to account for detection reliability without fundamentally redesigning the entire autonomous vehicle system.
Solution Approach 2:
The patent transforms the detection output by adding uncertainty parameters alongside the detected object information. Instead of only providing object location and classification, the system now provides detection confidence levels and uncertainty metrics, changing the parameter set to include reliability information that influences subsequent navigation decisions.
2Measurement precision
If the system incorporates uncertainty data from machine-learned models, then prediction accuracy and motion planning reliability improve, but computational complexity and processing requirements increase
Solution Approach 1:
The patent performs uncertainty quantification as a preliminary step during the object detection phase, before motion planning begins. By calculating uncertainty metrics upfront using the machine-learned detection models, the system prepares reliability information in advance, avoiding the need for complex real-time uncertainty calculations during critical motion planning and control phases.
Solution Approach 2:
The system implements feedback loops where uncertainty information from detection models is fed back into the motion planning process. This feedback mechanism allows the motion planner to adjust its decisions based on detection reliability, and the results can be used to refine future detection and uncertainty estimation, creating a continuous improvement cycle.
Data Source
AI summary
Systems and methods are provided for detecting objects of interest. A computing system can input sensor data to one or more first machine-learned models associated with detecting objects external to an autonomous vehicle. The computing system can obtain as an output of the first machine-learned models, data indicative of one or more detected objects. The computing system can determine data indicative of at least one uncertainty associated with the one or more detected objects and input the data indicative of the one or more detected objects and the data indicative of the at least one uncertainty to one or more second machine-learned models. The computing system can obtain as an output of the second machine-learned models, data indicative of at least one prediction associated with the one or more detected objects. The at least one prediction can be based at least in part on the detected objects and the uncertainty.