• state estimation
• hybrid legged-wheeled robot navigation
• path planning
• navigation among movable obstacles
• reconfigurable navigation planning

Autonomous navigation planning for robots is a widely studied research area. The capability to navigate with robustness, safety, and speed in cluttered and human-unsafe environments, such as in typical disaster scenarios, further motivate the need for autonomous planning algorithms. Various robots in the literature primarily use two main methods of navigation, namely wheeled and legged locomotion. There exist various mapping algorithms that fuse data from multiple sensors to create maps that are used as inputs to navigation planners to determine collision-free, low-cost motion paths from an initial pose to a goal position. Hybrid legged-wheeled robots, such as the CENTAURO (primary used in this thesis), are capable of performing agile, reconfigurable hybrid motions. Due to their high degrees of freedom, agility, and reconfigurabilty, the search space of possible motion plans becomes very high dimensional, requiring heavy computations with subsequently larger computation times. This problem is further compounded in cluttered environments, which is sparse on free space and simply driving using a fixed configuration wheeled motion is not possible. Furthermore, the clutter makes it very difficult to compute safe steps via legged-only motion. This thesis, focuses on the work done in the development of an Agile Reconfigurable Navigation Suite for the CENTAURO robot. We first present a state estimation algorithm, adapted from two state-of-the-art algorithms, before moving on to the main focus of this thesis, namely, the agile reconfigurable legged-wheeled motion planning. To traverse cluttered environments, we focus on two methods or modes. They are the footprint reconfiguring wheeled-only motion and the obstacle pushing motion. To that extent, we first present two footprint reconfiguring planner prototypes that allow us to push the limits of possible wheeled-only motion in cluttered areas with very low plan computation times. This is achieved by reducing the robot motion search space, allowing for quick collision-free footprint determination through 2D map image pixel search, and adapting popular grid-based planning algorithms, namely the A* and Theta*, to find low-cost footprint, modifying plans to reach the goal position. We demonstrate the effectiveness of the newly presented planners through image based simulations, demonstration on the simulated robot, and experiments on the real robot in cluttered environments created in the lab. We, then, present a new local object push planning algorithm. Its goal is to plan the pushes to be performed on movable obstacles that obstruct trajectories planned by the aforementioned wheeled-only motion planners. We parametrize the movable objects, use conservative region partitioning and image pixel geometry, to determine push sequences consisting of in-place single legged pushes, drive-through single legged pushes and drive through double legged pushes. The planned push sequence moves the obstacle to a location where the robot either has sufficient space to go around the obstacle or the pushed obstacle is in a new configuration that allows the robot modify its footprint and clear the obstruction so as to rejoin the original planned trajectories. This new modality of obstacle pushing allows the robot to clear blocked entry and exits to passages further improving the ability of the robot to traverse cluttered environments. All the developed methods are experimentally validated using either or both simulated and real-world robots.