Evolution Strategies Combined with Central Pattern Generators for Head Motion Minimization during Quadruped Robot Locomotion Cristina P. Santos, Miguel Oliveira, Herm´ ınia Mendes, Manuel Ferreira and Lino Costa Abstract— In autonomous robotics, the head shaking induced by locomotion is a relevant and still not solved problem. This problem constraints stable image acquisition and the possibility to rely on that information to act accordingly. In this article, we propose a movement controller to generate locomotion and head movement. Our aim is to generate the head movement required to minimize the head motion induced by locomotion itself. The movement controllers are biologically inspired in the concept of Central Pattern Generators (CPGs). CPGs are modelled based on nonlinear dynamical systems, coupled Hopf oscillators. This approach allows to explicitly specify parameters such as amplitude, offset and frequency of movement and to smoothly modulate the generated oscillations according to changes in these parameters. Based on these ideas, we propose a combined approach to generate head movement stabilization on a quadruped robot, using CPGs and an evolution strategy. The best set of parameters that generates the head movement are computed by an evolution strategy. Experiments were performed on a simulated AIBO robot. The obtained results demonstrate the feasibility of the ap- proach, by reducing the overall head movement. I. INTRODUCTION Visually-guided locomotion is important for autonomous robotics. However, there are several difficulties, for instance, the head shaking that results from the robot locomotion itself that constraints stable image acquisition and the possibility to rely on that information to act accordingly. The motion of quadruped, biped and snake-like robots, for instance, with cameras mounted in their heads, causes head shaking.This kind of disturbances, generated by locomotion itself, makes it difficult to keep the visual frame stable and, therefore, to act according to the visual information. Head stabilization is very important for achieving a visually-guided locomotion, a concept which has been suggested from a considerable num- ber of neuroscientific findings in humans and animals [16]. In this article, we aim to build a system able to minimize the head motion of a quadruped robot that walks with a walking gait. We propose a motion stabilization system for the head of an ers-7 AIBO quadruped robot. Basically, head motion is set such to generate the movement opposed to the one induced by the locomotion itself. Cristina Santos, Miguel Oliveira, Herminia Mendes and Manuel Ferreira are with Industrial Electronics Department, School of Engineering, University of Minho, 4800-058 Guimaraes, Portugal cristina@dei.uminho.pt, mcampos@dei.uminho.pt, hmendes@dei.uminho.pt, mjf@dei.uminho.pt Lino Costa is with Production Systems Department, School of Engineering, University of Minho, 4710-057 Braga, Portugal lac@dps.uminho.pt Several similar works have been proposed in literature [4], [7], [6], [5]. But these methods consider that the robot moves according to a scheduled robot motion plan, which imply that space and time constraints on robot motion must be known before hand as well as robot and environment models. As such, control is based on this scheduled plan. Other works have successfully achieved gaze stabilization [5], that consists on image stabilization during head movements in space. The overall of the gaze stabilization approaches can be divided into two types of techniques. One uses specific hardware, like accelerometers and gyroscope to estimate the 3D posture of the head, and complex control algorithms to compensate the oscillations. The use of inertial information was already proposed by several authors [5], [14], [15]. Typically this kind of techniques is used in binocular robot heads, where gaze is implemented through the coordination of the two eye movements. Most of the approaches are inspired in biological systems, specifically in the human Vestibular-Ocular Reflex (VOR). In robots with fixed eyes, the fixation point procedure is achieved by compensatory head or body movements, based on multisensory information of the head. In this work, we propose a combined approach to generate head movement stabilization on a quadruped robot, using Central Pattern Generators (CPGs) and Evolution Strategies (ESs) [17], [18]. We intend to use a head controller, based on Central Pattern Generators (CPGs), that generates tra- jectories for tilt, pan and nod head joints. CPGs are neural networks located in the spine of vertebrates, able to generate coordinated rhythmic movements, namely locomotion [11]. These CPGs are modelled as coupled oscillators and solved using numeric integration. These CPGs have been applied in drumming [1] and postural control [3]. This dynamical sys- tems approach model for CPGs presents multiple interesting properties, including: low computation cost which is well- suited for real time; robustness against small perturbations; the smooth online modulation of trajectories through changes in the dynamical systems parameters and phase-locking between the different oscillators for different DOFs. In order to achieve the desired head movement, opposed to the one induced by locomotion, it is necessary to ap- propriately tune the CPG parameters. This can be achieved by optimizing the CPG parameters using an optimization method. The optimization process is done offline according to the head movement induced by the locomotion when no stabilization procedure was performed.