A parameter search for a Central Pattern Generator (CPG) for biped walking is difficult because there is no methodology to set the parameters and the search space is broad. These characteristics of the parameter search result in numerous fitness evaluations. In this paper, nonparametric estimation based Particle Swarm Optimization (NEPSO) is suggested to effectively search the parameters of CPG. The NEPSO uses a concept experience repository to store a previous position and the fitness of particles in a PSO and estimated best position to accelerate a convergence speed. The proposed method is compared with PSO variants in numerical experiments and is tested in a three dimensional dynamic simulator for bipedal walking. The NEPSO effectively finds CPG parameters that produce a gait of a biped robot. Moreover, NEPSO has a fast convergence property which reduces the evaluation of fitness in a real environment.
Unlike animals and humans who are very adept at push recovery, humanoid robots push recovery is difficult for its high dimensional, non-linear, and hybrid features. Existed research results such like Capture points, provide several methods to recover from the push. However when a high magnitude push applies to the humanoid, existed methods are not enough to recover the robot. Towards this problem, Continuous Steps method is proposed to solve this problem in this paper.We present simulation of a simple humanoid that can recover from a high magnitude push by using continuous steps. Future work involves extending the modeling to arbitrary direction pushes and applying the method to the real humanoid robots.
This paper introduce a state classification method for detecting falling of biped robot. The method uses a Support Vector machine (SVM) to classify the state. The input vector for the SVM are a magnitude of acceleration, a position of center of pressure (CoP) in x and z axis, and tilt angles of torso relative to x and z axis. The input vector is based on sensor data that is measured from accelerometer and force sensing resistor (FSR) sensor. Training of the classifier is done in off-line and the trained classifier is used to classify the state of the biped robot in on-line. The method was verified in a 3D dynamics simulator and showed it could classify falling state within 0.01 second.
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.