epicon

What the framework does

EPICON is a LiDAR based aerial exploration package built for unmanned aerial vehicles that map unknown spaces. The authors describe it as a method for avoiding revisitation on point cloud maps, which is the central problem in exploration: a drone should cover new ground instead of flying over areas it has already scanned. The work was accepted by IEEE Robotics and Automation Letters in 2026, and the source code was released in August 2026. It builds directly on two earlier open source projects, EPIC and FALCON, which the authors credit for reducing the engineering effort needed to reach a working baseline. The package targets both simulation and real world flights, and the paper reports that the framework cuts redundant revisits and shortens total exploration time when compared with prior state of the art methods. The repository ships with a citation block for the RA-L paper so researchers can reference the method correctly. Because it is a research release, the emphasis is on reproducing the published experiments rather than on production deployment. The code is written in C++ and depends on ROS Noetic and Ubuntu 20.04, which sets a clear floor for anyone who wants to try it. The authors also thank the STAR lab and the teams behind MARSIM and FALCON for open sourcing the building blocks.

How it avoids revisitation

The core idea is to guide coverage using a path plan on the point cloud map so the vehicle does not return to spaces it has already observed. Revisitation is the usual failure mode in exploration: without explicit guidance, a planner tends to loop back through cleared areas, wasting battery and time. EPICON structures the flight around coverage paths derived from the map, which keeps the agent moving toward frontier regions that still need scanning. The README frames the result as eliminating revisitation redundancy rather than merely lowering it, and the paper evaluates three test scenarios called cave, garage, and city. Those three maps are the ones named in the launch instructions, so a new user can reproduce the exact setups the authors measured. The simulation environment is built on the GPU version of MARSIM, which means a machine without a GPU may need to modify the simulator before it runs. This dependency is called out plainly in the documentation so users do not waste time on a setup that will not build. The design goal stays close to the original EPIC baseline: fast UAV exploration, but with coverage path guidance layered on top to stop the agent from retracing its own path. The source lives in the zikangyuan-epicon repository on GitHub and is implemented in C++, with the license recorded as GPL-3.0 in the project metadata.

Building and running it

The documented build path uses a catkin workspace. You create a directory, clone the repository, and run catkin_make to compile the package. After the build, the example launch file is garage.launch, and you can swap garage for cave or city to match the three evaluated scenarios. Maps themselves are not bundled in the repo: the README points to simulation maps hosted by the EPIC and EDEN projects on Google Cloud, and you place the downloaded point clouds under MARSIM/map_generator/resource. That separation keeps the repository small but means a first run requires fetching external data before anything will launch. The authors also list a credits section that names EPIC, MARSIM, and FALCON as the projects that supplied the baseline framework, the simulator, and the large scale evaluation harness. If you plan to extend the work, those three repos are the places to look first. The documentation is short and research oriented, so expect to read the paper for the algorithmic detail and treat the README as a build and run guide rather than a full tutorial. ROS Noetic and Ubuntu 20.04 remain the supported environment. The source lives in the zikangyuan-epicon repository on GitHub and is implemented in C++, with the license recorded as GPL-3.0 in the project metadata.

Reproducing the paper results

Reproducing the published numbers means using the three named maps and the documented launch flow, then comparing exploration time and revisitation against the baselines the paper cites. The citation block in the README gives the exact bibliographic entry, including the journal, volume, number, and pages, so a writeup can reference it without hunting through IEEE. Because the framework is a drop in extension of EPIC, the comparison with that baseline is the most direct one to make. The acknowledgments note that EPIC and FALCON provided key insights and reduced implementation effort, which tells you the codebase sits close to those two projects in structure. Real world experiments are mentioned in the introduction as validation, but the README focuses on the simulation launch for day to day use. A researcher who wants the physical flights will need the hardware stack implied by the LiDAR setup and is expected to adapt the simulator notes accordingly. The release date of August 2026 places this among recent RA-L work, and the YouTube video linked from the README shows the exploration behavior for readers who want a quick visual check before building. The source lives in the zikangyuan-epicon repository on GitHub and is implemented in C++, with the license recorded as GPL-3.0 in the project metadata.

Editorial conclusion

The code is written in C++ and released under the GPL-3.0 license, with the paper and source hosted on the project repository.