This repository is the implementation of the MIP, MIP-PRH and MIP-SRH models for the Min-Max Rooted Tree Cover (MMRTC) problem and their corresponding planners for the graph-based multi-robot coverage path planning problem from the following paper:

Jingtao Tang and Hang Ma. "Mixed Integer Programming for Time-Optimal Multi-Robot Coverage Path Planning with Heuristics." IEEE Robotics and Automation Letters (Aug. 2023). [paper], [video], [project]

Please cite this article if you use this code for the multi-robot coverage path planning problem.

pip install -r requirements.txt

optional if you don't want to run solver.py for MIP optimization. Pre-run model solutions are provided in directory 'data/solutions'.

Please refer to [Gurobi] for the installation. (they have trial and academic licenses)

The MCPP problem is reduced to the MMRTC problem and then solved with the STC algorithm. Please refer to the paper for more details.

python solver.py [-h] [--solver_cfg SOLVER_CFG] [--alpha ALPHA] [--beta BETA] [--warm_start WARM_START] istc

• Required:
  • istc: the instance name stored in directory 'data/instances'.
• Optional:
  • --solver_cfg SOLVER_CFG: path to the Gurobi configuration file. (see 'data/cfgs' for reference)
  • alpha ALPHA: parameter of Parabolic Removal Heuristics (PRH). Will solve the MIP-PRH model if specified.
  • beta BETA: parameter of Subgraph Removal Heuristics (SRH). Will solve the MIP-SRH model if specified.
  • --warm_start WARM_START: type of warm-startup for the model optimization. Use 'RTC' for the original MIP model and 'MST' for MIP-PRH and MIP-SRH.

A simple routine to create random MMRTC instance.

• if map is provided, then a terrain with uniform terrain weight of 1 is generated, encoded by:
  • obstacle vertex: black pixel, rgb=(0,0,0)
  • free vertex: white pixel, rgb=(1,1,1)
  • root vertex: red pixel, rgb=(1,0,0)
• otherwise, an empty terrain with random weights will be generated.

python instance_maker.py [-h] [--map MAP] name

• Required:

  • name: the instance name in the format of '[grid width]x[grid height]-[Characteristics]-k[# of roots]'.
    • If no map is provided, the generated instance is a [grid width]x[grid height] empty terrain with [# of roots] subtrees (or robots) and randomized terrain weights.

• Optional:

  • --map MAP: path to the map to create the instance.

The MCPP planners with simulation, including MFC, MSTC$^*$ and MIP (the method in this paper).

python planner.py [-h] [--method METHOD] [--istc_sol_name ISTC_SOL_NAME] [--scale SCALE] [--dt DT] [--write WRITE] istc

• Required:
  • istc: the instance name stored in directory 'data/instances'.
• Optional:
  • --method METHOD: planner type choose from {MFC, MSTC*, MIP}.
  • -istc_sol_name ISTC_SOL_NAME: MIP solution name stored in the directroy 'data/solutions'. (only required when planner type is MIP)
  • --scale SCALE: the canvas scaling factor for visualization.
  • --dt DT: delta time of simulation.
  • --write WRITE: is writing the simulation as MP4. (ffmpeg lib is required)

• The floor-medium instance using the MMRTC solution from MIP-SRH model

MIP-MCPP is released under the GPL version 3. See LICENSE.txt for further details.