Article Preview
Top1. Introduction
Robotics has long been trying to emulate human characteristics, one of this characteristics is the capacity of balance and posture recovery. Humans are capable of this due to three systems that work together: The vestibular system, the visual system and the proprioception (Macpherson & Horak, 2012). This systems can be emulated by a robot with a series of sensors, like gyroscopes and accelerometers for the vestibular system, cameras for visual system and touch sensors for proprioception. The acquired information is processed in the cerebellum and the cerebral cortex, determining the motor responses in order to adjust the posture. The motor responses are translated into balance strategies.
In the literature about robot balancing strategies, there are three basic ones, which are used to recover the position after an unknown force is applied to them, these are the ankle, hip and step strategies.
On (Ono, Sato, & Ohnishi, 2011) an ankle strategy is used together with the knee joint to maintain balance on impacts bigger than the ones that can be tolerated by the ankle strategy on its basic implementation. On (Yasin, Huang, Xu, & Zhang, 2012) a step strategy is used calculating where to move the foot based on the use of trajectories, rescue and alternative sequences stored on memory whenever an unbalanced motion is detected. (Hyon, Osu, & Otaka, 2009) implements a hybrid strategy based on the ankle and hip strategies for both dynamic and static balancers. (Humphrey, Hemami, Barin, & Krishnamurthy, 2010) also uses hip and ankle hybrid strategies for robots that deal with perturbations on the surface in which they are supported, in this case, implementing into the robot a system similar to the human vestibular system.
There are previous works that attack the problem of balancing robots using models inspired by the human vestibular system (Mergner, Schweigart, & Fenell, 2009) (Patane, Laschi, Miwa, Guglielmelli, Dario, & Takanish, 2004) (Chen, Liu, & Chen, 2010), however, the scope of this work only provides the sensory part of the vestibular system (ampulla and otoliths) and use Bayesian models to make corrections to the position of the robot.
(Patane, Laschi, Miwa, Guglielmelli, Dario, & Takanish, 2004) tried to design and implement a vestibular system for robot heads in order to hold it's gaze for simulating movements using vestibulo-ocular reflex. To achieve the goal, they designed a 3-axis vestibular system and information fusion for process the information from the vestibular system and create motor commands to eye muscles.
In (Chen, Liu, & Chen, 2010) a vestibular system model for AIBO ERS-7 robots was constructed, their model consists of 3 parts: a method for computing poses, a low pass filter to smooth acceleration and a pattern for differentiation of surfaces. These three steps are performed with mathematical calculations.
In (Mergner, Schweigart, & Fenell, 2009) the use of an array of sensors inspired by the vestibular system and the somatosensory system, resulted in a human-like behavior, however nothing is mentioned about the calculation of poses.
Using the same approach used in (Cervantes, Rodriguez, López, Ramos, & Robles, 2013) and (Torres, Jaime, & Ramos, 2013) we strive to design and implement a synthetic vestibular system, said system should react in a human-like way. This document is structured as follows. The second section describes the proposed model for the vestibular system. The third, describes the implementation phase of the cognitive architecture, the sensor array, the robot movements and the communication between the three of them. The fourth section explain the results obtained from the architecture, sensors and communication, also the failed experiments with the robot. Finally, the fifth section tries to explain the need of another robot to show the balance strategies in real-time.