Insect walking and a walking-controlled robot

responsibility for wording of article: Akira Takashima (OIST

Walking insects can cope with a rough terrain. They can also use their legs for sensing and manipulating objects. The control system for these activities is composed of what appear to be low-precision circuit elements - neurons. However, this control system is robust enough to tolerate noises and demonstrate the ability of self-organization. The control system has a distributed processing structure, relying not only on sensory feedback (feedback loops from environmental interaction) but also on internal feedback.

Machines that move on land usually use wheels or caterpillar tracks; however, their use is generally restricted to a flat ground. In contrast, insects can move on natural land that has not undergone artificial preparation. Insects can walk around rugged and sandy places, climb tree branches, and track over obstacles of various sizes and shapes on the ground. Insect legs are also used in complex operations such as swimming, food gathering, grooming, and moving objects. In insects and crustaceans, their leg actions are controlled by relatively small brains. The processing speed and accuracy of neurons constituting their brains are vastly inferior to those of modern computers. However, the leg control system must be reliable and robust and must continue to function even if some legs are missing. It must also function with a low energy consumption. Therefore, an artificial walking system based on the insect control mechanism may be interesting not only to biologists but also to engineers.

At first glance, complex behavioral control would seem to require a complex brain. However, it has been shown that the information obtained from the use of physical properties of the body and its interactions with environment can drastically simplify the controlling mechanisms. Research on insect walking supports this idea. An artificial neural network (ANN) model called Walknet has been constructed using results from research on stick insect locomotion. This ANN model is conceptually close to biological systems. In addition, it can perform learning using online and offline information and is robust to input noises.

There are 2 basic problems related to walking control. The first is about control mechanisms for individual leg movements and the second is those of coordination between legs. A control mechanism that generates periodic stepping motions is related to the first problem. This leg control system behaves similarly to a relaxation oscillator, and the state transition (transition between the swing phase and stance phase) is dependent on a threshold on the basis of the leg position. The second problem is known to involve several local coordination mechanisms adjusting leg movements to implement the most appropriate gait.

Control of legs

Walking requires switching the leg movement between the stance phases, where the leg is in direct contact with the ground and propels the body forward, and the swing phase, where the leg is lifted up and returned to the position for starting the next stance phase. A very simple feed-forward neural network control model has been proposed on the basis of our knowledge on the flexible swing movement of the legs of living organisms. The control systems in operation during the ground phases are not well known. Feedback from the movement speed and force at each joint is considered to be important. However, this factor has not been examined using the robot.

The transition from a stance phase to a swing phase is an important issue that requires the control of individual legs and coordination between adjacent legs. Inter-leg coordination is particularly important for avoiding a fall of the body. Experiments on insects and crustaceans suggest that the leg position is important for lifting the legs (i.e., transition from a stance phase to a swing phase) and that tactile and pressure sensors are important in transition from a swing phase to a ground phase. Walknet that is based on research on stick insect locomotion does not use devices such as central pattern generators (CPGs) and central oscillators. This indicates that central oscillators are not necessarily required for slow walking. However, in fast-moving insects such as cockroaches, CPGs and actuators with elastic elements are thought to be necessary to cause the appropriate mechanical vibrations.

Coordination between legs

Another important problem related to coordination is inter-leg coordination of the transition from a stance state to a swing state. The conventional solution for walking with 6 legs is based on a rule of moving 2 sets of 3 legs alternately. However, this fixed walking pattern is only applicable to situations where there are no outside disturbances. More detailed studies on insects and crayfish have shown that there are many localized rules governing the movement of adjacent legs. Walknet uses a distributed adaptive gait controller on the basis of these rules. Interestingly, the receptors for gravity found in, for example, mammals for assisting with body stability are not found in insects. Instead of these, walking insects use output from feedback loops controlling each leg joint.

How can the knowledge of insect locomotion control be applied to robots? The first step is simulation research. Another approach is to construct a real robot.

Further Reading

昆虫ミメティックス Insect Mimetics(2008),針山孝彦,下澤楯夫,pp.878-884

updating of the site
Copyright (C) 2018 Neuroinformatics Unit, RIKEN Center for Brain Science