Research in robot locomotion can be separated into few groups. The dominated research is based on a high level modeling approach, like ZMP control that incorporate inverse kinematics techniques, which suffers from the problem of high dimensionality, delays, and the requirement of perfect knowledge of the robot and environment. Taken inspiration from neuroscience, with the attention on biological inspired locomotion for robot control [1]. These controllers do not require perfect knowledge of the robot's dynamics as with classical strategies and have shown great promise in terms of robustness, simplicity and adaptivity. Biological inspired locomotion controllers are based on simple circuits that are built from sensor-neurons, motor neurons, and inter-neurons [2] [3]. Neurophysiological studies associate rhythmic movements with the oscillation activity of a particular type of neurons, called neural oscillators [4] [5]. These oscillators can produce rhythmic activity without sensory inputs and even without any central inputs. But the sensory information is indispensable for walking because it allows shaping of the rhythmic patterns in order to interact with the environment [6]. However, sensory information is mainly used to adapt the controller in the event of changes and perturbations. Neurophysiologists have shown that biological controllers like Central Pattern Generators (CPG) have adaptation properties due to neural plasticity mechanism [4] [7]. With inspiration from neurobiology, Ijspeert et al. proposed different models for rhythmic movements control [3]. The neural reflexive walking controller, proposed by F. W¨org¨otter is one of the few that have been tested on a real bipedal robot [2]. Our work aims to produce a robust biological inspired neural controller for biped walking, based on CPG with a rhythmic neuron proposed by Rowat and Selverston [5]. We show therefore how to adapt against an external perturbation force by phase resetting or by behavior adapting. This paper is organized as following. Section 2 presents the principles of the neural controller based on the model of rhythmic neurons, which is able to generate CPG-like patterns. The three layers of the CPG used in bipedal control are presented. A coupling circuitry for walking is proposed. Next, two approaches to deal with for external perturbation is presented. The last section gives a conclusion and details of further developments.