ForgeScan is a library for developing voxelized reconstructions and next-best view selection policies.
Autonomous robotics systems rely on digital reconstructions of environments and objects to make decisions. Whether navigating in an evolving space or interacting with a new object, a system must first observe its surroundings.
ForgeScan breaks this process into three components: data representations, with focus on voxel data structures; next-best view selection algorithms; and distance measurements, from sensors depth cameras.
This project is still in early development and subject to major changes. This is still primarily a research tool in its alpha version. An updated library based on this will be released soon.
Project documentation is available here.
Docker files are provided with a description and documentation of the packages required for developing this code on linux systems1.
The following collaboration diagram depicts how the Manager class, containing the reconstruction
data representation and implements view selection policies, interacts with simulated depth cameras
and overall system.
Footnotes
-
The code will compile on windows if the same packages are installed. ↩