HARP

Fly, Drive, Reconfigure: A Modular Reconfigurable Aerial-Ground Platform for Field Operations

Li-Yu Lo*, Yanbaihui Liu*, Chengchuan Shu, Tyler Harris, Jonathan Ryan, Boyuan Chen

* Equal contribution

General Robotics Lab, Duke University

Website · Video · Paper (arXiv) · PDF · Hardware · Mapping · Planning

HARP (Heterogeneous Aerial Robotic modules Platform) combines aerial scouts, fly-drive rovers, and task-specific payloads. Modules fly independently and physically assemble into a cooperative ground vehicle for payload transport. This repository contains the hardware designs, ROS 2 mapping tools, and energy-aware dynamic programming (DP) route planner accompanying the research.

Abstract

Heterogeneous robot teams distribute complementary capabilities across specialized agents, but their physical roles and capacities typically remain fixed throughout a mission. We present HARP, a Heterogeneous Aerial Robotic modules Platform in which independently deployable aerial robots physically reconfigure to compose their capabilities for field operations. HARP comprises sensor-equipped scouts, fly-drive rover modules, and task-specific payload modules. Scouts map the environment and inform an energy-aware planner that jointly selects routes and air-ground mobility modes. Rover and payload modules fly independently across terrain that constrains ground travel, then autonomously assemble into a cooperative ground vehicle for energy-efficient payload transport. Motivated by environmental sampling in remote and difficult-to-traverse regions, we evaluate HARP through field experiments spanning sensing, planning, reconfiguration, air-ground mobility, payload transport, and task execution. We further conduct module-level deployment tests on the Greenland Ice Sheet toward future autonomous missions. HARP demonstrates how heterogeneous robot teams can adapt not only their actions, but also how their physical capabilities are composed during a mission.

System overview

The paper integrates terrain sensing, mobility planning, and physical reconfiguration into a field-operation pipeline:

1. Map: A scout surveys the environment to provide terrain information.
2. Plan: An energy-aware DP planner jointly selects routes and aerial or ground mobility modes.
3. Reconfigure and transport: Rover and payload modules transition between independent flight and assembled ground travel.
4. Execute: A task-specific payload performs field operations, demonstrated with a drilling mechanism for subsurface sampling.

Hardware

Fly-drive rover module. Aerial deployment and ground mobility.	Drilling payload module. Aerial deployment with a subsurface sampling mechanism.

See the rover design and drilling payload design for CAD assemblies, part inventories, and reference images.

Repository

Component	Contents	Documentation
CAD/	Rover and drilling payload designs, assembly inventories, and images.	Hardware guide
mapping/	ROS 2 tools for 2.5D mapping, exploration, PCD export, and browser-based map editing.	Mapping and map editor guide
path_planner/	Air/ground DP planner, point-cloud terrain conversion, example maps, configuration, and regression tests.	Planner setup and examples

Getting started

• Explore the hardware: Start with the CAD guide and the assembly inventories for each module.
• Build or edit a terrain map: Follow the mapping guide. Mapping consumes registered point clouds and odometry from a separately running FAST-LIO instance.
• Run the planner on an example map: Follow the planner setup and examples, which include Python dependencies, a sample command, visualization outputs, and point-cloud conversion.

Citation

If you use HARP in your research, please cite the paper:

@misc{lo2026flydrivereconfigure,
  title         = {Fly, Drive, Reconfigure: A Modular Reconfigurable Aerial-Ground Platform for Field Operations},
  author        = {Li-Yu Lo and Yanbaihui Liu and Chengchuan Shu and Tyler Harris and Jonathan Ryan and Boyuan Chen},
  year          = {2026},
  eprint        = {2609.28887},
  archivePrefix = {arXiv},
  primaryClass  = {cs.RO},
  url           = {https://arxiv.org/abs/2609.28887}
}

Acknowledgement

This work is supported by a Dean’s Research Venture Fund from Duke Nicholas School of the Environment, DARPA TIAMAT program under award HR00112490419, and ARO under award W911NF2410405.