Enhancing adaptability with local reactive behaviors for hexapod walking robot via sensory feedback integrated central pattern generator

Highlights

• A two-layered CPG-based single-leg controller is devised to generate tripod movement.

• Two local reactive mechanisms are proposed to deal with terrain disturbances.

• A locomotion control framework is proposed and verified for hexapod robot.

Abstract

Local reactive behaviors endow animals the ability to exhibit agile and dexterous performance when traversing challenging terrains. This paper presents a novel locomotion control method based on the central pattern generator (CPG) concept for hexapod walking robot with local reactive behavior to cope with terrain irregularities. Firstly, a two-layered CPG-based single-leg controller is developed to generate the rhythmical movement for each leg executing tripod walking. The Van der Pol oscillator is employed on the high-layer to construct a coupled CPG network which serves as a phase regulator (PR) to produce rhythmic signals with prescribed phase relations amongst neurons. On the low-layer, an auxiliary linear converter (LC) transforms these signals into the desired joint trajectories. Subsequently, by embodying the proprioceptive sensing and external tactile information as the sensory feedback, two typical local reactive mechanisms including the elevator reflex and searching reflex are achieved by virtue of on-line adjusting the coupling scheme of the PR and the coefficients of the LC. A locomotion control framework for hexapod walking robot is further established by combining the single-leg controller with a finite state machine to allocate swing/stance commands for individual joints in dealing with terrain perturbations. The effectiveness of the proposed method has been verified through both virtual model simulation and experiments on a physical hexapod platform.

Introduction

Achieving agile and robust walking performance as presented in vertebrates and arthropods becomes increasingly attractive for legged robots when traversing complicated terrains in recent years [1], [2], [3]. Due to the inherent redundancy among the joints and the complexity in foot-ground interaction, tackling the motion coordination of multi-degrees of freedom (MDoF) and multi-axial movement is regarded as a challenge in locomotion control of legged robots [4]. The relevant biological observations and experimental results provide abundant inspirations for scientists and engineers to deal with this issue. By refining the essence of the neuro-mechanics, a serials of legged prototypes such as RHex [5], Zebro [6], Foldable Hexapod [7], PhantomX [8], HITCR-II [9], Octopus-III [10] and AMOS [11] have been successfully developed, exhibiting astonishing adaptive behaviors in varied surfaces. Recently the desert ant-inspired 3D printed hexapod robots [12] have been devised and provided an affordable alternative in outdoor exploration due to the distinctive adaptability in unknown environment [13], [14]. With regard to traversing rough terrains, a position feedback-based adaptive locomotion control method has been proposed in [15] for affordable hexapod walking robot to improve the mobility in typical unstructured terrains (i.e., stairs, irregular blocks). Nonetheless, the underlying mechanism on how animals maneuver tissues, organs as well as limbs to produce energetically efficient and dexterously movements when interacting with the unstructured environment still remains unsettled.

Recent neuroscience research has found that the central pattern generator (CPG), a group of neural circuit located in the spinal cord of vertebrates or in relevant ganglia in invertebrates, can produce rhythmic and quasi-periodic movement in absence of sensory feedback or higher regulation signals from the brain-stem level [16]. Meanwhile, the neurobiology studies [17], [18], [19] have revealed that the local reactive behaviors via muscles, tissues and organs with tactile sensing or vestibular feedback plays an irreplaceable role for terrestrial creatures when coping with unconventional environment. These biological observations and findings inspire both engineers and roboticists to resolve specific locomotory tasks for legged robots. Lewinger et al. [20] proposed a biologically inspired leg controller to generate adaptive stepping actions in a hexapod walking robot. The proposed controller contains three sensory-coupled neuron circuits by collecting the angle and load information of each joint, exhibiting a motion combination including the protraction of the thorax-coxa, the levation of the coxa-trochanter as well as the extension of the femur–tibia. Rutter et al. [21] extended this work by developing a neural network-based single-leg controller with local sensor feedback to steer a cockroach-like robotic leg. By appropriately inhibiting or exciting the local neural circuits and adjusting the associated neuron thresholds, a serials of adaptive reflex transitions were obtained on the single-leg platform. Similar results could be found in [22] wherein the movement of a 3-DoF leg is steered by the proposed joint controller that embodies simplified linear muscle models. The strategy based on bio-inspired local reactive behavior was also implemented in [23] that the reactive climbing performance for hexapod walking robot AMOS-II is conducted via the central pattern generator (CPG) network integrated with reflex neurons. The robot could negotiate obstacles with varied height, displaying a comparable climbing behaviors as observed in cockroaches. The reflex mechanism had also been cooperated with the CPG network to successfully manipulates a limbless robot for improving adaptability in unstructured environment with rapid response in presence of external stimuli [24]. Kimura et al. [25] developed a CPG-based controller for the quadruped prototype Tekken-II in which the corrective stepping reflex together with the crossed flexor reflex induced by sensory feedback was implemented on the swing leg to maintain the balance when dynamically walking on irregular terrain in outdoor environment. Wang et al. [26] also achieved stable and robust hopping for a biped robot in uneven surface by elaborately adding sensory feedback path to the original CPG circuits. Espenschied et al. [27] proposed a distributed locomotion controller incorporating with local leg reactive behaviors including step reflex, elevator reflex as well as searching behaviors for a insect-like hexapod robot to enhance the walking adaptation when negotiating complicated terrains such as irregular, slatted and compliant surface.

Although these bio-inspired reflex control methods are widely and fruitfully applied in locomotion control of legged robots, it still remains a challenge for locomotion control of legged robots that how to generate suitable and efficient reactive behaviors comparable to natural performance of animals. Generally, constructing the CPG network for locomotion control is a trade-off between the complexity of the neural network topology and the diversity of the resulting gait patterns. From the perspective of synchronization of complex network, the rhythmical signals with stable phase relationships that are appropriate to generate diverse gait patterns for legged locomotion depends on the collective dynamical behaviors of the CPG network with specific coupling scheme amongst neurons. However, the sophisticated topology of the CPG network will inevitably reduce the mathematical tractability, which further decreases the possibility of straight-forward parametrical synthesis of the CPG-based controller for locomotory tasks in steering limb/limbless movement. Traditional numerical methods such as evolutionary searching [28], empirical tuning [29] as well as other learning approaches [30], [31] are applied to obtain the appropriate model parameters of the neural network. Constructing an ideal CPG network capable of providing abundant dynamical behaviors that results rhythmic locomotion patterns meanwhile preserving tractability is a long-term goal for the CPG-based locomotion control for legged robots especially in cope with unstructured terrain profiles.

Aiming at improving the adaptability of hexapod walking robot in unstructured environment, this paper presents a novel CPG-based locomotion controller with sensory feedback to generate local reactive behaviors for single-leg in dealing with irregular terrain. To fully leverage the merit of the CPG-based approach in producing rhythmic signals with stable phase relationships among neurons, a coupled neuron network integrating with proprioceptive sensing and external tactile information to exhibit conventional leg movement in tripod gait and two typical adaptive reactive behaviors. The main contributions of this paper are summarized as follows.

• A two-layered CPG-based single-leg controller is developed to generate reference trajectory for hexapodal walking in conventional tripod gait. On the high-level, the CPG network composed of three coupled Van der Pol oscillators is constructed as the phase regulator (PR) to produce rhythmic signals with prescribed waveforms. These signals are transformed into the desired joint trajectories for a single-leg by using a linear convertor (LC) on the low-level.

• Two typical local reactive behaviors including the elevator reflex and searching reflex are achieved by adjusting the coupling scheme of the PR and the coefficients of the LC. The former reflex mainly deals with irregular terrain (i.e., step obstacles) to avoid stumbling while the latter treats sunk terrain (i.e., gap, ditches or holes) to search feasible foothold.

• A comprehensive locomotion control framework for hexapod walking robot by combining the single-leg controller with two reflex mechanisms is established by using a finite state machine (FSM) to schedule individual legs in swing/stance phase during walking. The effectiveness of the proposed method is demonstrated through the simulation on a virtual hexapod model and the experiments on physical hexapod robot.

The remainder of this paper is organized as follows. Section 2 briefly revisits the neurobiological basis of the motor system in animals. Section 3 presentsthe development of the CPG-based single-leg controller with sensory feedback. Reactive behavior generation is detailed in Section 4 followed by the establishment of locomotion control framework for hexapod robot in Section 5. The effectiveness of the proposed controller is validated via both simulation and experiments in Section 6. The paper ends with conclusions in Section 7.