Navigation Framework for Humanoid Robots Integrating Gaze Control and Modified-univector Field Method to Avoid Dynamic Obstacles Jeong-Ki Yoo and Jong-Hwan Kim Abstract— This paper proposes a navigation framework for humanoid robots, which integrates gaze control and modi- fied univector field-based path planning to cope with moving obstacles. To make navigation robust, obstacles are modeled according to their relative velocities and positions. Moreover, partial evaluation values for gaze control architecture are also considered for modifying their virtual size and moving trajec- tory. In addition, gaze control architecture is proposed, which estimates the size of local map confidence area, self-localization error, surrounding obstacles and obstacle-free distance against those obstacles in the local map. The proposed framework is verified through computer simulations by using a developed simulator for HanSaRam-VIII. I. I NTRODUCTION Most of humanoid robot researches have focussed on walking issues [1], [2]. From the biologically inspired ap- proach to the dynamic model-based approaches [3], [4], they have focussed on generating the dexterous walking pattern. Due to the remarkable improvements in hardware and walk- ing pattern generation for humanoid robots, researches are now expanding to various other fields of robotics, such as navigation, vision perception, task processing in complex environments, etc [5]–[7]. In the navigation aspect, various navigational concepts for wheeled robots, such as heuristic search algorithm, dynamics-based random state search approach and force- based algorithm, have been applied to humanoid robots. A* algorithm was applied to footstep planning [8], and vision-guided footstep planning in dynamic environment was performed [9]. Among various search algorithms, rapidly- exploring random trees approach was also applied for motion planning of humanoid robots [10]. Since it estimates and explores through sampled states maintaining its dynamic constraints, it can rapidly derive quasi-optimal path. Univec- tor field method using virtual obstacle concept for footstep planning was proposed, which is expanded in this paper [20]. Integration with vision processing algorithms was performed in many researches of humanoids, which focussed on in- creasing autonomy and obstacle avoidance [11]. Along with these approaches, researches related to gaze control are also important not only for stabilizing vision images, but also for efficient navigation [12]. It plays a major role in gathering surrounding information [13]. Information The authors are with the Department of Electrical Engineering, KAIST, 335 Gwahangno (373-1 Guseong-dong), Yuseong-gu, Daejeon 305-701, Republic of Korea (e-mail: {jkyoo, johkim}@rit.kaist.ac.kr). theory-based approach was proposed in [14]. Even though these are closely dependent on path planning issues, integra- tion aspect has not been considered as a broadly concerned research topic so far. This paper proposes a navigation framework, which in- tegrates gaze control and modified univector field method [20]. Since the univector field method is robust for real-time applications, it is modified to deal with moving obstacles efficiently by adopting the proposed dynamic virtual obstacle and velocity modification scheme. Instead of using a simple duplicated virtual obstacle, dynamically resizing and moving obstacle integrated with partial evaluation functions for gaze control is proposed. In addition, its move-to-goal univector field function is modified to pass by via points and arrive at a goal position with arbitrary arriving posture angle. Gaze control system integrates four partial evaluation functions for local map confidence area size, self-localization error from covariance matrix, environmental status of obstacles, and obstacle-free distance of local map. These are also used for deriving the size and length of the path of virtual obstacle. Through this interlinked structure, the performance of the whole framework is expected to have synergy effects for avoiding collision with moving obstacles. Verification of the proposed scheme is performed through computer simulations with a model of a small-sized humanoid robot, HanSaRam- VIII (HSR-VIII), developed at the Robot Intelligence Tech- nology (RIT) Laboratory, KAIST since 2000 [15]. This paper is organized as follows. In Section II, gaze control architecture focusing on partial evaluation functions is described. Section III explains the modification scheme for univector field method using dynamic virtual obstacle and velocity modification approach. Section IV describes an integration of the proposed framework. Section V presents simulation environment and results, and finally conclusions follow in Section VI. II. GAZE CONTROL ARCHITECTURE This section describes the gaze control architecture based on four types of partial evaluations and their integration scheme. A. Partial Evaluations for Gaze Control In case of navigation in a complex environment, various types of information, such as relative distances and speed The 2010 IEEE/RSJ International Conference on Intelligent Robots and Systems October 18-22, 2010, Taipei, Taiwan 978-1-4244-6676-4/10/$25.00 ©2010 IEEE 1683