المفاهيم الأساسية
A domain-invariant underwater navigation approach that maximizes data collection of objects of interest while avoiding obstacles, without relying on localization.
الملخص
The paper introduces UIVNAV, a novel underwater navigation method that aims to maximize the collection of data on objects of interest (OOI) while avoiding obstacles, without relying on localization.
The key aspects of the approach are:
Generating an Intermediate Representation (IR) from the robot's camera images, which includes depth information and segmentation of the OOI. This IR is designed to be domain-invariant, allowing the navigation policy to be applied across different environments and OOIs without retraining.
Training a domain-invariant navigation policy using imitation learning, where a human expert labels the IR frames with desired yaw and pitch changes to guide the robot towards larger areas of the OOI and avoid obstacles.
The authors demonstrate the effectiveness of UIVNAV in simulation across various underwater environments, including oyster reefs and rock reefs. Compared to a complete coverage method and a random walk approach, UIVNAV is able to survey on average 36% more of the OOI while traveling 70% less distance. The authors also present a real-world deployment of UIVNAV on a BlueROV underwater robot in a pool, successfully navigating and surveying a bed of oyster shells.
The key advantages of UIVNAV are its domain-invariance, the ability to efficiently gather data on OOI without relying on localization, and its potential to be integrated with higher-level global exploration and coverage algorithms.
الإحصائيات
The robot using UIVNAV surveys on average 36% more oysters compared to the complete coverage method when traveling the same distances.
UIVNAV travels on average 70% less distance compared to the complete coverage method while collecting 29% less oyster data.
اقتباسات
"UIVNAV chooses to visit the areas with larger area sizes of oysters or rocks with no prior information about the environment or localization."
"A robot using UIVNAV compared to complete coverage method surveys on average 36% more oysters when traveling the same distances."