Complex Motion Planning for NAO Humanoid Robot Walaa Gouda1 and Walid Gomaa1;2 1Computer Science and Engineering, Egypt-Japan University for Science and Technology (E-JUST), Alexandria, Egypt 2Faculty of Engineering, Alexandria University, Alexandria, Egypt Keywords: Whole Body Motion, Humanoid Robot, Complex Dynamic Motions , Robot Kinematics , Recognized Object. Abstract: In this paper, we introduce an integrated approach that enables a humanoid robot to plan and robustly execute whole body motions including stepping over, climbing up or down obstacles as well as climbing up straight staircase using only onboard sensing. Reliable and accurate sequence of motions for humanoid robots op- erating in complex indoor environments is a prerequisite for robots to fulfill high level tasks. The design of complex dynamic motions is achievable only through the use of robot kinematics. Based on the recognized object from the robot database, using the robot camera, a sequence of actions for avoiding that object is ex- ecuted. As demonstrated in simulation as well as real world experiments with NAO humanoid, NAO can reliably execute robustly whole body movements in cluttered, multi-level environments containing objects of various shapes and sizes. 1 INTRODUCTION cution (Oßwald et al., 2011). However, there are reasons that explain why hu- Robots have always been a subject of curiosity for manoid robots aren’t used frequently in practical ap- both generalists and technologists alike. Humanoids, plications. One of these reasons is that humanoids are robots with multiple degrees of freedom, have be- expensive in cost, as they consist of complex pieces come popular research platforms as they are con- of hardware and are manufactured in small numbers sidered to be the future of robotics. The human (Maier et al., 2013). Also, many researchers apply like design and locomotion allow humanoid robots to navigation algorithms that represent a humanoid us- perform complex motions. This includes balancing, ing wheels instead of legs, but the limitation of this walking, access different types of terrain, standing model is that it does not respect all the navigation ca- up, step over or onto obstacles, reaching destinations pabilities of humanoid robots and therefore more ap- only accessible by stairs or narrow passages, and to propriate approaches are necessary for navigation in navigate through cluttered environments without col- cluttered and multi-level scenarios (Maier et al., 2013; liding with objects. These abilities would make hu- Hornung et al., 2010; Gouda et al., 2013). manoid robots ideal assistants to humans, for instance In the beginning, humanoid robotics research fo- in housekeeping or disaster management (Graf et al., cused on specific aspects like walking, but now cur- 2009; Maier et al., 2013). rent systems are more complex. Many humanoid Autonomous obstacle avoidance by stepping over, robots are already equipped with full body control onto/down the obstacle, climbing stairs with hu- concepts and advanced sensors like stereo vision, manoid robots is a challenging task, since humanoids laser, auditory and tactile sensor systems which is the typically execute motion commands only inaccurately essential condition to deal with complex problems, (Graf et al., 2009; Maier et al., 2013; Shamsuddin such as walking and grasping. Motion planning is et al., 2011). This is due to the fact that humanoids a promising way to deal with complex problems, as possess only a rough odometry estimate; they might planning methods allow the flexibility of different cri- slip on the ground depending on the ground friction, teria satisfaction. The design of complex dynamic and backlash in the joints might occur. Additionally, motions is achievable only through the use of robot the observations of their small and light weighted sen- kinematics, which is an analytical study of the motion sors are inherently affected by noise. This all can lead of the robot manipulator (Maier et al., 2013; Kucuk to uncertain pose estimates or inaccurate motion exe- and Bingul, 2006; Gienger et al., 2010). 402 Gouda W. and Gomaa W.. Complex Motion Planning for NAO Humanoid Robot. DOI: 10.5220/0005051904020409 In Proceedings of the 11th International Conference on Informatics in Control, Automation and Robotics (ICINCO-2014), pages 402-409 ISBN: 978-989-758-040-6 Copyright c 2014 SCITEPRESS (Science and Technology Publications, Lda.) ComplexMotionPlanningforNAOHumanoidRobot More specifically, robot kinematics provide the ping over and climbing up/down obstacles as well as transformation from the joint space, where the kine- climbing up straight staircase in a 3D environment, matic chains are defined, to the Cartesian space, shown in figure 2. Relying only on the robot on- where the robot manipulator moves, and vice versa board sensors, joint encoders, an efficient whole body (Kofinas, 2012). Robot kinematics are quite useful, motions planning perform safe motions to robustly because it can be used for planning and executing navigate in challenging scenes containing obstacles movements, as well as calculating actuator forces and on the ground as shown in figure 2. Our approach torques. Robot kinematics can be divided into for- determines the appropriate motion that consists of a ward and inverse kinematics. The forward kinematics sequence of actions according to the detected obsta- refers to the use of the kinematics equations of the cle using monocular camera and bumper sensors. As robot to compute the position of the end effector from demonstrated in practical experiments with a NAO specified values of the joint parameters (Kucuk and humanoid and in a series of simulations experiments Bingul, 2006). On the other hand the inverse kine- using Webots for NAO humanoid robot, which is a matics refers to the use of the kinematics equations simulation software for modeling, programming and of a robot to determine the joint parameters that pro- simulating robots (Cyberbotics, 2014), our system vide a desired position of the end effector. It is easy leads to robust whole body movements in cluttered, to see why kinematics is required in any kind of com- multi-level environments containing objects of vari- plex motion design (Kucuk and Bingul, 2006; Kofi- ous shapes and sizes. nas, 2012). The relationship between forward and inverse kinematics is illustrated in figure 1. Balancing meth- ods rely on the ability to calculate the center of mass of the robot, which is constantly changing as the robot moves. Finding the center of mass is made possible only if the exact position and orientation of each part of the robot in the three dimensional space is known Figure 2: The simulated environment. (Graf et al., 2009). The remainder of this paper is structured as fol- lows. Related work is discussed in the Section II. Section III describes the humanoid robot, also mo- tion design and object learning phase used for exper- imentation are described in this section . Section IV Figure 1: The schematic representation of forward and in- illustrates the robustness and accuracy of our motion verse kinematics. planning approach in experiments. Finally, Section V concludes the paper. Humanoid robots performing complex motions tasks need to plan whole body motions that satisfy a variety of constraints. As the robot must maintain 2 RELATED WORK its balance, self-collisions and collisions with obsta- cles in the environment must be avoided and, if pos- Humanoid motion planning has been studied inten- sible, the capability of humanoid robots to step over sively in the last few years. For instance the ap- or onto objects, navigate in multi-level environment proach presented by (Oßwald et al., 2011) enabled an needs to be taken into account. These constraints and equipped NAO humanoid robot with a 2D laser range the high number of degrees of freedom of the hu- finder and a monocular camera, to autonomously manoid robot make whole body motion planning a climb up spiral staircases. While (Hornung et al., challenging problem (Graf et al., 2009). The main 2010) presented a localization method for NAO hu- goal of whole body balancing motion is to generate manoid robots navigating in arbitrary complex indoor and stabilize consistent motions and adapt robot be- environments using only onboard sensing. Also the havior to the current situation (AldebaranRobotics, approach developed by (Nishiwaki et al., 2002) al- 2014). lowed NAO to climb single steps after manually posi- In this paper, an integrated whole body motion tioning the robot in front of them without integrating planning framework has been developed. The frame- any sensory information to detect the stairs. Footstep work enables the robot to robustly execute whole actions plan to climb staircases consisting of three body balancing sequences of actions, including step- steps with HRP-2 is introduced by (Chestnutt et al., 403 ICINCO2014-11thInternationalConferenceonInformaticsinControl,AutomationandRobotics 2007). While (Samakming and Srinonchat, 2008) presented a technique for climbing stair robot using image processing technique besides reducing the pro- cessing time. While (Maier et al., 2013) designed motion, called T-step, that allows the robot to make step over actions, as well as parameterized step onto and step down ac- tions. The authors in (Yoshida et al., 2005) investi- gated a dynamic pattern generator that provides dy- namically feasible humanoid motion including both locomotion and task execution such as object trans- portation or manipulation. While (Shahbazi et al., 2012) introduced a learning approach for curvilinear bipedal walking of NAO humanoid robot using pol- icy gradient method. Their proposed model allows Figure 3: Aldebaran NAO H25. for smooth walking patterns and modulation during walking in order to increase or decrease robot speed. Algorithm 1: Navigate through the environment. A suitable curvilinear walk, very similar to human or- 1 move three steps forward dinary walking, was achieved. 2 stop moving Furthermore an approach to whole body motion 3 pitch down NAO head by 30◦ planning with a manipulation of articulated objects 4 switch to NAO lower camera such as doors and drawers is introduced in (Burget 5 look for obstacle et al., 2013).