Most of the mobile robots nowadays essentially move on the ground without wall-climbing capability. The "City-Climber" robots developed at the City College of New York (CCNY) are able to move on virtually any kind of smooth or rough surfaces and can carry a relative large payload. Unlike the conventional ground mobile robots operating in 2D space, and UAVs in 3D space, the City-Climber robots operate in constrained 3D space, i.e., its action space is confined within planar surfaces while the sensing space is 3D, facilitated by the freedom of motion on ground, walls, and ceilings. This attribution provides both opportunities and challenges. One of the benefits is that the City-Climber can take vantage positions on a ceiling or wall to gain better view of the scene. The challenges lie in the fact that most of the existing planning/control methods for multi-robot systems are no longer valid and it demands new framework to deal with this hard yet previously unexplored research domain involving wall-climbing robots. The PI proposed to develop a general framework and provide a theoretical foundation to deal with planning, control and coordination problems for a group of heterogeneous wall-climbing robots and ground robots operating in the constrained 3D space of urban environments. The issues to be studied in this thrust include: abstraction of world models for complex urban environments; computational efficiency and data structure for the world abstraction and representation; dynamic view planning and motion planning for a multi-robot team in constrained 3D space; a framework integrating the discrete world representation, high level planning and lower level control primitives; case studies of the framework using several exemplar tasks (i.e., self-deployment, surveillance coverage, formation control of multi-robots, and target tracking), and the experimental validation of the proposed methods and algorithms. Being an integral part of the career development plan, the proposed education activities include: integration of research into curriculum, mentored research experience for students, improving the educational pipeline with coherent education and outreach activities, enhancing the robotics extension programs targeted for K-12 education.
The research activities of this CAREER project consist of two thrusts: 1) to conduct basic research on the planning, control and coordination of a group of heterogeneous robots (i.e., wall-climbing robots, ground robots and UAVs) operating in constrained 3D space; 2) to develop new wall-climbing robots and use them as research tools to experimentally validate proposed methods. The project is distinguished by the multi-scale (from individual robots to multi-robot systems), wide-spectrum research (for different canonical tasks such as 3D motion planning, 3D mapping and localization, multi-robot exploration and formation control) around a unified theme (i.e., advancing mobile robot to 3D). The City-Climber robots take advantage of the merits of aerodynamic attraction and vacuum suction, thus make good balance between strong adhesion force and high mobility. Since the City-Climber robots don’t require perfect sealing, the robots can operate on both smooth and rough surfaces and can carry large payload. These attributions enable the City-Climber robots gain bird-eye view of the environments from ceiling and avoid occlusions. The City-Climber robots not only have tremendous applications but also open a new special domain of research. One of our research efforts is to develop a theoretical framework for collaborative 3D map construction in indoor environments using multiple robots and multi-modal sensing (e.g., cameras, laser range scanners, and RGB-D sensors). The intellectual merits of the research include a systematic method and general framework to fuse 3D laser scans obtained from ground robots and wall-climbing robots on the ceiling, which uses an overhead camera on the wall-climbing robot and solves the Perspective Three Point (P3P) Problem to obtain the geometric relationship among the robot team, and then uses the obtained transformation matrix as a good initial estimate to refine the laser scan registration using improved Iterative Closest Point (ICP) algorithm. The experimental validation (as shown in Figure 1) indicates that the method doesn’t need camera/scanner calibration and the ICP algorithm is guaranteed to converge fast, resulting in a composite 3D laser scan map of the environment. The framework is extended to obtain large scale map by dual robots in stop-n-go fashion as shown in Figure 2. In order to improve accuracy and reduce processing speed for 3D mapping, we have developed several algorithms including: 1) a method to cluster noisy range images into planar regions by means of patch-based sampling; 2) a novel scan matching algorithm that applies the positional relationship of the objects as the features for scan matching so that the amount of the matching pairs is much smaller than the raw 3D laser points; 3) a fast and robust polygon-based 3D registration algorithm whose flow chart is shown in Figure 3. As one of the pioneering groups investigating RGB-D sensors in robotics applications, we have developed visual odometry and 3D mapping algorithms using RGB-D camera that is so fast that it is suitable for real-time implementation. Another research direction is on theoretical investigation of multi-robot exploration of unknown indoor environments that can be modeled as an undirected graph as shown in Figure 4. We analyze mathematically the behavior of different algorithms (i.e., Multi-Robot Depth First Search, and Flooding Algorithm) obtaining their main properties and specifying the bounds on the exploration time and traversed distance. Other research efforts include the investigation of 3D path planning using mixed integer linear programming (MILP), examination of the multi-solution phenomenon of the P3P problem and introduction of two methods (a deterministic and a probabilistic method) that guarantee a unique solution to the P3P problem. The research results are disseminated by means of CCNY open source software package that is accessible by ROS.org community (www.ros.org/wiki/ccny-ros-pkg), and eleven journal papers, two book chapters, and more than forty peer-reviewed conference papers, including five candidates for Best Paper/Best video awards in robotics conferences. In addition, a minimal viable product (MVP) of City-Climber robot targeted specifically for the building inspection and non-destructive testing (NDT) industries are developed for real world applications and broader impact. As an integral component of the career development plan, the PI have made tremendous effort in creating/enhancing an educational "pipeline" to attract high school students to STEM disciplines, to retain and timely graduate minority undergraduates, and to mentor graduate students with MS and Ph.D. research. Some exemplary education and outreach activities include: serving as the faculty mentor of CCNY Robotics Club and CCNY Autonomous Vehicle Design Club, mentoring CCNY teams to participate various competitions, offering robotics class at CCNY STEM Institute in summers, mentoring high-school teams and serving as a judge for the NYC FIRST robotics competitions since 2003, mentoring summer interns and REU students at CCNY Robotics Lab, making project demonstration at many STEM events such as the world science festival (WSF2012) as shown in Figures 5. The students under PI’s supervision have performed extremely well and become winners of many prestigious fellowships and national competitions.
