Experience Based Surface Discernment by a Quadruped Robot

Abstract

Abstract-The task of autonomous surface discernment by an AIBO robotic dog is addressed. Different surface textures (plywood board, thin foam, short carpet, shag carpet) as well as different inclines (0 and 10 degrees) are considered. Using a genetic algorithm, gaits are designed which allow the robot to traverse each of these surfaces in an (approximately) optimal fashion. Frequency domain analysis of actuator readings from individual leg joints is performed for data collected using each gait on each surface type. It is found that the spectral content of these signals is significantly dependent on the characteristics of both the gait in use and the surface being walked upon. Using tap-delay Adaline neural networks to integrate actuator readings from 15 independent joints into a set of models of different gait/surface experiences, an algorithm is designed which uses these experiences to yield high classification rates across surface transitions and with low latency.

Rigidity-Based Surface Recognition for a Domestic Legged Robot

“…Unlike when the robots walked on one example per domestic surface type in [5,8,9,10,13,15], AIBO run on multiple examples per type to gather samples in a more generic manner in this paper. The intraclass variability in the dataset were higher in this way and the interclass correlation had a higher chance tough these properties were not studied.…”

“…While proprioceptive sensors (e.g potentiometers) were researched in [1,3,6,8,9], the ERS-7 model has a force sensor in each leg joint. The previously mentioned papers combined all joints in the sensor readings, however, the author found real discriminative power for the hip joints of the hind legs during the initial experiments.…”

“…Various locomotion options and sensor combinations have been examined in the past works: wheeled [11,12,16,17], quadruped [3,4,8,9,10,13], six-legged robots [15]. Although some studies analyzed a few sensors [9,11] or the body oscillations [18], this paper does not go into this direction, but the focus is on the classifier model building and the evaluation to provide a better interpretation of the past results from machine learning point of view.…”

“…Although some studies analyzed a few sensors [9,11] or the body oscillations [18], this paper does not go into this direction, but the focus is on the classifier model building and the evaluation to provide a better interpretation of the past results from machine learning point of view. Most reviewed studies evaluated their models properly [1,3,8,9,10,12,13], but some did not employed (cross-)validation to get a better estimate of the real accuracy [6,11,15]. This paper uses a four-legged Sony ERS-7 robot and the feature vector is composed by the readings of four sensors to train the models.…”

“…The reviewed literature solved the surface recognition challenges with machine learning methods, but Holmstrom et al applied experience based models to the problem [9]. Specific gaits were evolved by a genetic algorithm for a Sony AIBO robot to walk on plywood board, thin foam, short carpet and shag carpet.…”