Abstract
There is still no artificial life with the same adaptability and flexibility as animals. Although artificial life will advance by acquiring nervous systems similar to those of animals, its role and mechanisms remain unknown. The authors have developed a quadruped robot with a bio-inspired gait generation method to realize robots that can behave like animals. The method could generate gaits using pulse-type hardware neuron models (P-HNMs). However, characteristics of the P-HNMs had large scatterings, and the robot could maintain gaits only a few cycles. This chapter explains the method and P-HNM integrated circuits (ICs) developed to improve P-HNMs' characteristics. In addition, dynamic simulations with a simplified method and discussions of the methods and ICs are provided. Although the proposed methods are simple, they could actively generate gaits using interactions between the body and the environment. Therefore, the methods will lead to the realization of a quadruped robot with flexible adaptability.
TopIntroduction
Artificial life is being studied in various fields, broadly categorized as soft, hard, and wet, and is becoming increasingly interdisciplinary (Aguilar, Santamaría-Bonfil, Froese, & Gershenson, 2014; Habib, 2011; Habib, Watanabe, & Izumi, 2007). The main topic of this chapter is control methods for legged robots, but since these methods are bio-inspired, they apply also to soft and wet fields.
Modern legged robots can perform advanced movements and are used in many applications (Biswal & Mohanty, 2020). For example, some of them can patrol in response to their surroundings using optical devices and other equipment. However, even though their mobility is already comparable to that of animals, none of them can act as autonomously as animals. One of the reasons is that responding autonomously to myriad situations is an arduous task for modern robot control methods. Legged robots need to complete the task as soon as possible. These robots achieve their robustness through advanced control systems that use many sensors (e.g., Raibert, Blankespoor, Nelson, & Playter, 2008; Fankhauser, Bjelonic, Bellicoso, Miki, & Hutter, 2018). The computation costs required to develop a robot that can act autonomously as animals will be incomparable to current ones. Although robots will gradually improve as computers become more powerful, as we have seen in the past, there is a significant capability gap between robots and animals. This problem arises because the computer, the robot’s brain, has to perform all calculations to expand its capabilities.
On the other hand, animals behave autonomously with seeming ease. The significant difference between robots and animals is whether the brain processes all behaviour or not. For example, animals unconsciously generate respiration, chewing, walking, and so on (e.g., Marder & Bucher, 2001; Selverston & Ayers, 2006). Particularly, since walking is one of the most important movements for both legged animals and robots, the knowledge about the generating mechanism will solve the problems limiting current robot control methods. Quadrupeds are the most common of legged animals, and there are many studies on their walking behaviour. These studies have shown that they have several locomotion patterns (gaits) that they switch depending on the situation (Bhatti, Waqas, Mahesar, & Karbasi, 2017; McMahon, 1985; Taylor, 1985). Findings on the relationship between horses’ locomotion speed and oxygen consumption are well known (Hoyt & Taylor, 1981). In addition, neurophysiology experiments have provided insights into the relationships of the nervous system to gait generation (e.g., Duysens & Pearson, 1980; Grillner, 1975; Orsal, Cabelguen, & Perret, 1990). The theory that quadruped animals unconsciously generate gaits by interacting with the central pattern generator (CPG) and sensory inputs is currently widely accepted (Bellardita & Kiehn, 2015; Frigon & Rossignol, 2006; Grillner & Zangger, 1979). Despite many discussions on animals’ gait generation mechanisms, much remains unknown (e.g., Arshavsky, Deliagina, & Orlovsky, 2016; Delcomyn, 1980).
Key Terms in this Chapter
FSR402: A force-sensing resistor. The authors used FSR402 with a voltage divider to use as a pressure sensor. The detailed characteristics are available online at: https://www.interlinkelectronics.com/fsr-402 .
Pulse-Type Hardware Neuron Model (P-HNM): A kind of artificial neuron model that emulates biological neurons’ functions by analogue electrical circuits. There are other artificial neuron models by mathematical equations.
Pulse Width Modulation (PWM): A method the microcontroller outputs voltages. Outputs a voltage by alternately turning on / off, and by integrating the output voltage, it can generate any voltage between on and off. The authors used this method to vary the output from the microcontroller in increments of approximately 0.8 mV.
KRS-2552RHV ICS: A common servomotor that can be controlled by serial or PWM mode. The authors controlled them with PWM mode using Arduino DUE.
CoppeliaSim: A dynamic simulator that supports several application programming interfaces (APIs) and remote APIs. The CoppeliaSim also supports four different dynamics engines: the Bullet physics library, the Open Dynamics Engine, Vortex Studio engine, and the Newton Dynamics engine. The authors simulated the quadruped robot model using the Newton Dynamics engine.
Arduino DUE: A single board microcontroller using ARM CPU. It has 12 PWM output pins and 12 analogue inputs. Although it also has two digital-to-analogue converters (DACs), the authors used the PWM output pins and the peripheral circuits because the number of the DAC is not enough for the quadruped robot system.
Gait: Phase differences between the legs during locomoting. Walk, trot, and gallop gait are typical gaits in quadrupedal locomotion. For example, in the walk gait, LF, LH, RF, and RH move in 90° phase difference, and in the trot gait, the diagonal legs move in 180° phase difference in synchronization.