Resumen de Dexterous robotic motion planning using intelligent algorithms
César Augusto Arismendi Gutiérrez
The fundamental purpose of robots is to help humans in a variety of difficult tasks, enabling people to increase their capabilities of strength, energy, speed, memory, and to operate in hazardous environments and many other applications. Service robots, more precisely mobile manipulators, incorporate one or two robotic arms and a mobile base, and must accomplish complex manipulations tasks, interacting with tools or objects and navigating through cluttered environments. To this end, the motion planning problem plays a key role in the ahead calculation of robot movements to interact with its world and achieve the established goals. The objective of this work is to design various motion planning methods in order to improve the autonomy of MANFRED-2, which is a mobile robot fully developed by the Robotics Lab research group of the Systems Engineering and Automation Department of the Carlos III University of Madrid. Mobile robots need to calculate accurate paths in order to navigate and interact with objects throughout their surrounding area. In this work, we have developed motion planning algorithms for both navigation and manipulation. The presented algorithms for path planning are based on the Fast Marching Square method and include a replanner with subgoals, an anytime triangular planner, and a nonholonomic approach. The replanner with subgoals starts by generating a smooth and safe global path with the Fast Marching Square method. Then, this path is divided into multiple subpaths separated by equidistant nodes (defined by topological or metric constraints). After that, the obstacles information is progressively added and modifications are made only when the original path is unreachable. The most important advantage with respect to similar approaches is that the generated sub-paths are always efficient in terms of smoothness and safeness. Besides, the computational cost is low enough to use the algorithm in real-time. The anytime triangular planner, such as “Anytime” algorithms, quickly finds a feasible but not necessarily optimal motion plan which is incrementally improved. One important characteristic that this type of algorithms must satisfy is that the path must be generated in real-time. The planner relies on the Fast Marching Square method over a triangular mesh structure. Different variants are introduced and compared under equal circumstances that produce different paths in response time and quality, which leads us to an additional consideration. As in the field of benchmarking it is becoming increasingly difficult to compare new planners approaches because of the lack of a general benchmarking platform, improvements to existing approaches are suggested. Finally, the nonholonomic approach is presented. It is based on the Fast Marching Square method and its application to car-like robots. In order to apply the proposed method, a three dimensional configuration space of the environment is considered. The first two dimensions are given by the position of the robot, and the third one by its orientation. This means that we operate over the configuration space instead of the bi-dimensional environment map. Besides, the trajectory is computed along the configuration space taking into account the dimensions of the vehicle. In this way, it is possible to guarantee the absence of collisions. The proposed method is consistent at local and global scale because it guarantees a motion path solution, if it exists, and does not require global replanning supervision when a local trap is detected. Once a mobile robot has reached a goal location, it usually triggers the servomotors enclosed inside its robotic arm to manipulate a target. The manipulation algorithms presented in this work include the adaptation of trajectories, a planner with adaptive dimensionality, and an implementation of a dimensionality reduction approach inside a nuclear device. The adaptation of manipulation trajectories enables the robot to accomplish a task in different locations by using Evolution Strategies and forward kinematics. This approach avoids all the inconveniences that inverse kinematics imply, as well as the convergence problems in singular kinematic configurations. The planner with adaptive dimensionality reduces the complexity of high-dimensional path planning. First, a Rapidly-exploring Random Tree trajectory is generated using the full degrees of freedom of the robotic arm. Then, a geometry as a closed tube is built around the path line and the Fast Marching Square method is applied from start to goal using three dimensions inside the surface. The resulted three dimensional path is converted to full degrees of freedom with the inverse kinematics of the robot. The result is a smoother and safer path, visually more human friendly. Additionally, the search space is reduced, and therefore, also the planning time and the memory requirements. The application inside the nuclear device, similarly to the previous approach, reduces the degrees of freedom of the problem (but this time to two dimensions due to the mostly planar nature of the robot). The manipulation path is smooth and safe in an environment where safety must be the primarily objective. The motion planning algorithms have been tested in numerous experiments. The fast response of the methods allows its application in real-time tasks.
