Research On Terrain Classification For Robots Based On Bayesian Framework

Autonomous mobile robot is an intelligent robot with autonomous driving and autonomous planning ability,and it can work in a specific environment.Recognition and detection ability is the important condition,which makes robot move autonomously and steadily.The mobile robot must have the ability to identify and get through safely all kinds of unknown ground,so the ground classification has become a hot research topic.Traditional method based on radar and vision is widely used in the terrian identification and classification,but this method is easily influenced by illumination,surface covering,and is difficult to identify soft terrians,So this paper adopts the method based on vibration signal.Different from non-contact method,the vibration signal can reflect the bearing layer information of terrains,and is an important supplement to the visual.The research work will be carried out from three aspects: The first is to present the feature extraction methods;The second is the design of multiclass relevance vector machine classifier;The last is terrain classification for robots based on bayesian framework.The paper sets up a data acquisition experiment system.A mobile robot that installed a three-direction accelerometer and a microphone in z direction on the four wheel arms respectively is used to acquisite the vibration signals of wheel-terrain interaction to classify the terrains by traversing on sand,gravel,grass,soil and asphalt terrains and a single traverse with five different velocities.The experiment provided data support for the latter research of classification algorithm in chapter three to five.The original experimental data are divided into short data segment.Using Fast Fourier Transform(FFT)and Power Spectral Density(PSD)feature extraction method,feature of original vibration signal is extracted to obtain a feature sample set,which is used to study the post order terrain classification algorithm.To the problem that the terrain classification is essentially a multi classification,this paper presents terrain classification method based on multiclass relevance vector machine(M-RVM),and this method obtains probability output form of terrain classification results,which convenient to classify terrain by using filtering algorithm based on bayesian framework.Five kinds of single terrain and a single traverse are classified by M-RVM based on FFT andPSD features with five different velocities.To the problem that the traditional classification algorithm is only based on single observation sample,and considering the temporally coherent between continuous observation samples of robots moving process,the paper takes the classification of historical information into the Bayesian framework to forecast current terrain.Combining prediction with the current observation,the paper classifies the terrain of current sample.The state transition model of terrain classification system based on Markov chain methods is presented.Based on the FFT and PSD features,and selecting M-RVM as the classifier,the particle filter and kalman filter algorithm are used to classify artificial path(built of the 5 kinds of single terrain)and transition surface at five different velocities.The proposed algorithms have been validated by corresponding experiments.