The design of robust vision based robot navigation behaviors remains a challenge in mobile robotics as it requires a coherent mapping between a complex visual perception and its associated robot motion. This contribution proposes a framework to learn this general relationship from a small set of representative demonstrations in which an expert manually navigates the robot through its environment. Behaviors are represented by a dynamic system that ties the perceptions to actions. The state of the behavioral dynamics is characterized by a small set of visual features extracted from an omnidirectional image of the local environment. Recording, learning and generalization takes place in the product space of visual features and robot controls. Training instances are recorded for three distinctive behaviors namely corridor following, obstacle avoidance and homing. Behavioral dynamics are represented as Gaussian mixture models, parameters of which are identified from the recorded demonstrations. The learned behaviors are able to accomplish the task across a diverse set of initial poses and situations. In order to realize global navigation, the behaviors are coordinated via hand designed arbitration or command fusion schemes. The experimental validation of the proposed approach confirms that the acquired visual navigation behaviors in cooperation accomplish robust navigation in indoor environments.

This paper presents a novel approach for semantic classification of scenes and places with omnidirectional images. The objective of scene classification is to segment and classify different regions in the image whereas place recognition assigns a single category to the entire image. In scene classification image regions are classified into categories floor, vertical planar surfaces and isolated objects e.g. furniture. The semantic segmentation extracts multiple heterogeneous visual features at the superpixel-level that are labeled by randomized decision trees. The place recognition relies on a global image representation (GIST) and two local densely extracted shape and appearance representations(HOG, dense SIFT). A support vector machine predicts place categories such as room, corridor, doorway and open space from these visual features.

This paper proposes a new robot learning framework to acquire scenario specific autonomous behaviors by demonstration. We extract visual features from the demonstrated behavior examples in an indoor environment and transfer it onto an underlying set of scenario aware robot behaviors. Demonstrations are performed using an omnidirectional camera as training instances in different indoor scenarios are registered. The features that distinguish the environment are identified and are used to classify the traversing scenarios. Once the scenario is identified, a behavior model trained by means of artificial neural network pertaining to the specific scenario is learned. The generalization ability of the behavior model is evaluated for seen and unseen data. As a comparison, the behaviors attained using a monolithic general purpose model and its generalization ability against the former is evaluated. The experimental results on the mobile robot indicate the acquired behavior is robust and generalizes meaningful actions beyond the specifics presented during training.

This paper presents a novel approach for floor obstacle segmentation in omnidirectional images which rests upon the fusion of multiple classification generated from heterogeneous segmentation schemes. The individual naive Bayes classifiers rely on different features and cues to determine a pixel's class label. Ground truth data for training and testing the classifiers is obtained from the superposition of 3D scans captured by a photonic mixer device camera. The classification is supported by edge detection which indicate the presence of obstacles and sonar range data. The complementary expert decisions are aggregated by stacked generalization, behavior knowledge space or voting combination. The combined floor classifier achieves a classification accuracy of up to 0.96 true positive rate with only 0.03 false positive rate. A robust robot navigation is accomplished by arbitration among a reactive obstacle avoidance and a corridor following behavior using the robots local free space as perception.

The design of visual robotic behaviors constitutes a substantial challenge. It requires to draw meaningful relation ships and constraints between the acquired visual perception and the geometry of the environment both empirically and programmatically. This contribution proposes a novel robot learning framework to classify and acquire scenario specific autonomous behaviors through demonstration. During demonstration, robocentric 3D range and omnidirectional images are recorded as training instances of typical robot navigation situations pertaining to different contexts in multiple indoor scenarios. A programming by demonstration approach generalizes the demonstrated trajectories to a general mapping between visual features extracted from the omnidirectional image onto a corresponding robot motion. The approach is able to distinguish among different traversing scenarios and further identifies the best matching context within the scenario to predict an appropriate robot motion. As a comparison to context matching, the behaviors are trained by means of an artificial neural network and its generalization ability is evaluated against the former. The experimental validation on the mobile robot indicates that the acquired visual behavior is robust and generalizes meaningful actions beyond the specific environments and scenarios presented during training.

This paper introduces a new approach for detecting free space and obstacles in omnidirectional images that contributes to a purely vision based robot navigation in indoor environments. Naive Bayes classifiers fuse multiple visual cues and features generated from heterogeneous segmentation schemes that maintain separate appearance models and seeds for floor and obstacles regions. Pixel-wise classifications are aggregated across regions of homogeneous appearance to obtain a segmentation that is robust with respect to noise and outliers. The final classification utilizes fuzzy preference structures that interpret the individual classification as fuzzy preference relations which distinguish the uncertainty inherent to the classification in terms of conflict and ignorance. Ground truth data for training and testing the classifiers is obtained from the superposition of 3D scans captured by a photonic mixer device camera. The results demonstrate that the classification error is substantially reduced by rejecting those queries associated with a strong degree of conflict and ignorance.

This paper describes a novel approach for purely vision based mobile robot navigation. The visual obstacle avoidance and corridor following behavior rely on the segmentation of the traversable floor region in the omnidirectional robocentric view. The image processing employs a supervised approach in which the segmentation optimal with respect to the appearance of the local environment is determined by cross validation over 3D scans captured by a photonic mixer device (PMD) camera. The range data in the front view provides the seeds and validation data to supervise the appearance based segmentation in the omniview. Segmentation relies on histogram backprojection which maintains separate appearance models for floor, obstacles and background. A naive Bayes classifier predicts the occupancy of the robots local environment by fusing the evidence provided by different segmentations and models. The classification error is analyzed on ground truth data generated by a PMD camera and manually segmented scenes. The scheme is highly robust with respect to ambiguous and misleading visual appearances of obstacles and floor, thus enabling the robot to navigate safely in unstructured environments of diverse appearance, texture and illumination. The proposed vision algorithm and the navigation behavior demonstrate a robust performance in extensive robotic experiments across several hours of autonomous operation.