Multi-Robot Coverage Path Planning for 3D Reconstruction and Mapping of Complex Structures

  • Randa Almadhoun

Student thesis: Doctoral Thesis


Randa Almadhoun. Multi-Robot Coverage Path Planning for 3D Reconstruction and Mapping of Complex Structures. Ph.D. in Engineering Research Thesis, Khalifa University of Science and Technology, Abu Dhabi, United Arab Emirates, April 2021. Recent years have seen an increase in the use of autonomous Unmanned Aerial Vehicles (UAVs) for data gathering in a wide range of applications. Most of these applications require providing complete coverage of 2D areas or 3D structures utilizing the process of Coverage Path Planning (CPP) including area surveillance, industrial inspection, search and rescue, and forest _re monitoring. CPP is the process of generating an optimized path that guarantees complete coverage of the structure or environment of interest, in order to gather highly accurate information. Di_erent CPP approaches have been surveyed addressing various challenges depending on the environment dimensionality, and the structure complexity. Most of the surveyed CPP approaches utilizes single robots or multi-robot systems for wide areas coverage utilizing classical approaches. However, in many cases, generating e_cient coverage paths for di_erent large 3D environments using a team of robots remains an open research challenge. Therefore, generating an e_ective CPP method for various structure shapes and e_cient in computation time could be facilitated using hybrid approaches by combining Machine Learning (ML) with conventional CPP methods. CPP is one of the active research topics that could bene_t greatly from multi-robot systems employing ML approaches. This thesis aims to tackle the issue of CPP for exploration and mapping in 3D environments using multi-robot systems and employing ML approaches. In particular, three contributions are proposed towards this aim including (1) an Adaptive Search Space Coverage Path Planner (ASSCPP), (2) a Guided Next Best View (NBV) approach and (3) a Hybrid CPP (HCPP) approach. The _rst contribution utilizes a reference model and onboard sensors' noise models to generate coverage paths that are evaluated based on the travelled distance, turning angle and the quality of the 3D model. The second contribution generates a dense 3D reconstruction for an unknown structure by performing initial scan to improve coverage completeness and ight time. This approach generates viewpoints iteratively based on a novel adaptive nearest neighbour and a frontier sampling method. Then it selects the best viewpoint based on a proposed utility function that evaluates information theory, model density, travelled distance, and predictive measures of existing structure symmetries. The third contribution generates coverage paths for unknown 3D structures following a new hybrid approach which shows the versatility of combining the guided NBV method with a developed Long Short Term Memory (LSTM) approach to decrease the computational time and achieve high coverage. A set of experiments were conducted in a simulated and real environment using di_erent models to test the validity of the proposed approaches. Di_erent evaluation criteria were used to compare and verify the e_ectiveness of the proposed approaches.
Date of AwardMay 2021
Original languageAmerican English


  • Coverage Path Planning
  • Viewpoints Generation
  • 3D Reconstruction
  • Autonomous Exploration
  • Path Prediction
  • Deep Learning.

Cite this