Adnan Tahirovic · tahirovic@elet.polimi.it Gianantonio Magnani · magnani@elet.polimi.it
ARTIGO TÉCNICO
BEST STUDENT
AN EXTENSION TO ROUGH TERRAINS OF THE MPC/CLF MOBILE VEHICLE NAVIGATION APPROACH ABSTRACT Model Predictive Control (MPC) combined with the Control Lyapunov Function (CLF) optimization framework has been used for the navigation planning of indoor mobile robots moving in flat terrains. This approach guarantees the stability in the Lyapunov sense of the planned trajectory, provided that a proper navigation function is included in the CLF. In this paper, an extension of the combined MPC and CLF approach is proposed for navigation planning in outdoor rough terrains. The extension is based on a novel theoretical consideration. The proposed algorithm ensures obstacle avoidance as well as the selection of an appropriately traversable terrain by optimizing an objective function which considers the terrain roughness level along admissible paths.
I. INTRODUCTION Planetary explorations, search and rescue missions in hazard areas, surveillance, humanitarian de-mining, as well as agriculture applications such as pruning vine and fruit trees, represent possible fields of using autonomous vehicles in natural environments. The unstructured environment and the terrain roughness including dynamic obstacles and poorly traversable terrains pose a challenging problem for the autonomy of the vehicle. A nice overview of motion planning has been presented in [1]. The main focus of the early research stage was finding collision-free paths. In [2] the potential field approach for real-time obstacle avoidance was introduced while the concept of navigation functions was illustrated in [3]. The following work given in [4] included the general path planning problem using high d.o.f. manipulators. Also, the motion planning for mobile robots operating in a structured environment was discussed, dealing with local minima problem as well. Ge and Cui dealt with the problem of moving obstacles using the potential field method [5]. The research on motion planning evolved by adding the capability of taking into account the vehicle motion constraint within the well known dynamic window approach [6], [7]. This subject was extended to the high-speed navigation of a mobile robot in [8] by the global dynamic window approach, as the generalization of the dynamic window approach. A combination of the dynamic window approach with other methods yielded some improvements in long-term realworld applications [9]. Dubowski and Iagnemma extended the dynamic window approach to rough terrains introducing the vehicle curvature-velocity space. In this space the stability constraints of the vehicle, for instance expressed by limit values of the roll-over and side slip indexes, can be easily described. The given algorithm was also suitable for high speed vehicles and appropriate for real-time implementation [10]–[12]. The work presented in this paper was mainly inspired by the MPC/CLF framework (Model Predictive Control and Control Lyapunov Function) derived and explained in [13] and its application to the mobile robot navigation problem in flat terrains proposed in [14]. The proposed algorithm extends and adapts the MPC/CLF optimization framework from flat to rough terrains preserving its main property of guaranteed task completion. This means taht the framework uses the MPC/
[10] robótica
CLF control paradigm for navigation purposes providing a merge of a local and a global planning within a compact single framework. This gives the possibility of proving the guaranteed task completion using the stability concept of the MPC/ CLF framework since Lyapunov function consists of the navigation function that deals with global planning. The cost function that is locally minimized within the MPC horizon describes the level of roughness that should be estimated for all candidate paths. The level of roughness along a candidate path represents the information on how hard is to traverse this path by the mobile vehicle. The presented MPC/CLF scheme for rough terrains navigates the vehicle to follow less rough paths unlike those generated by the MPC/CLF for flat terrains. The main practical consequence of the selection of less rough terrain sections is the increase of the vehicle ability for high-speed maneuvers that do not cause unwanted effects such as sideslip and rollover. In accordance to the MPC optimization, any additional constraint can be imposed into the MPC/CLF navigation, such as those related to vehicle stability preventing from vehicle rollover and unnecessary sideslip. Unlike sample-based approaches where the optimization is inherently off-line, such as one variant of lattice roadmap paradigm [15], [16] where state lattices where created to represent differential constraints of the vehicle, the MPC/CLF method is suitable for online operations. In this paper, the analytic proof of the maximum task completion time and path length are also presented. In Section II, the MPC/CLF optimization scheme [13] and its application to the navigation planning for flat terrains [14] are reviewed. Section III explains theoretically the approach proposed to deal with different levels of terrain roughness, while Section IV gives analytical proof of the maximum time and length of the task completion. The simulation results and the conclusion are presented in Section V and VI.
II. PREVIOUS WORK USED IN THIS PAPER A. Dynamic Window Approach The idea of using a vehicle velocity space for the local obstacle avoidance appeared in [6]. It aims at optimizing an objective function dealing with