Deep Learning systems have proven to be extremely successful for image recognition tasks for which significant amounts of training data is available, e.g., on the famous ImageNet dataset. We demonstrate that for robotics applications with cheap camera equipment, the low image quality, however, influences the classification accuracy, and freely available data bases cannot be exploited in a straight forward way to train classifiers to be used on a robot. As a solution we propose to train a network on degrading the quality images in order to mimic specific low quality imaging systems. Numerical experiments demonstrate that classification networks trained by using images produced by our quality degrading network along with the high quality images outperform classification networks trained only on high quality data when used on a real robot system, while being significantly easier to use than competing zero-shot domain adaptation techniques.

As vision-based autonomous systems, such as self-driving vehicles, become a reality, there is an increasing need for large annotated datasets for developing solutions to vision tasks. One important task that has seen significant interest in recent years is semantic segmentation. However, the cost of annotating every pixel for semantic segmentation is immense, and can be prohibitive in scaling to various settings and locations. In this paper, we propose a region-based active learning method for efficient labeling in semantic segmentation. Using the proposed active learning strategy, we show that we are able to judiciously select the regions for annotation such that we obtain 93.8% of the baseline performance (when all pixels are labeled) with labeling of 10% of the total number of pixels. Further, we show that this approach can be used to transfer annotations from a model trained on a given dataset (Cityscapes) to a different dataset (Mapillary), thus highlighting its promise and potential.

This paper presents a recursive method for extracting planar surfaces from noisy range data. The method first transforms the range image into a so-called Enhanced Range Image (ERI) that encodes the local geometric information (surface normals) and global spatial information (coordinates) of the 3D range data. The ERI is then clustered into a number of homogenous groups called Super-Pixels (SPs). By treating the SPs as the nodes a graph is constructed. A new similarity function is proposed to compute the edge weights between the nodes, base on which the graph is recursively partitioned into two segments by the Normalized Cuts (NC) method until an exit condition is met. In this work, the exit condition is that each of the resulting segments is a plane or contains only one SP. After the partitioning process, neighboring planar segments are merged based on their spatial relationships. The recursive approach eliminates the need for a pre-specified segment number that is necessary in the existing NC based image segmentation methods. The ERI coding enhances object surfaces and edges while the effect of its sensitivity to surface normals is suppressed by the similarity function that takes into account the spatial information in computing the edge weights of the graph. The proposed method can be applied to navigation of mobile robots, symbolic map-building, and range data understanding. In this work the range data are captured from a 3D Time-of-Flight imaging sensor-the Swissranger SR-3000.

This paper presents a new method for extracting planar features from noisy range data. The method encodes the local geometric information (surface normals) and global spatial information (coordinates) of 3D data points into an Enhanced Range Image (ERI) which is then clustered into a number of homogeneous groups, called Super Pixels (SPs). The Normalized Cuts (NC) method is employed to the graph built on the SPs and groups the SPs into planar segments. The ERI coding enhances object surfaces and edges while its sensitivity to surface normals is suppressed by the NC measure that takes into account the spatial information of SPs in computing the edge weights of the graph. A binary matrix is constructed to represent the spatial and similarity relationships among the planar segments. We then employ a search algorithm on this matrix to merge homogenous planar segments. The proposed approach is compared with a representative plane segmentation method in various indoor environments and the results demonstrate the efficacy of the proposed method. In this paper, rang data are captured from a 3D imaging sensor-the Swissranger SR-3000.

This paper presents a new method for extracting object edges from range images obtained by a 3D range imaging sensor the SwissRanger SR-3000. In range image preprocessing stage, the method enhances object edges by using surface normal information; and it employs the Hough Transform to detect straight line features in the Normal-Enhanced Range Image (NERI). Due to the noise in the sensor's range data, a NERI contains corrupted object surfaces that may result in unwanted edges and greatly encumber the extraction of linear features. To alleviate this problem, a Singular Value Decomposition (SVD) filter is developed to smooth object surfaces. The efficacy of the edge extraction method is validated by experiments in various environments.

This paper describes a new approach to extract planar features from 3D range data captured by a range imaging sensor-the SwissRanger SR-3000. The focus of this work is to segment vertical and horizontal planes from range images of indoor environments. The method first enhances a range image by using the surface normal information. It then partitions the Normal Enhanced Range Images (NERI) into a number of segments using the Normalized-Cuts (N-Cuts) algorithm. A least-square plane is fit to each segment and the fitting error is used to determine if the segment is planar or not. From the resulting planar segments, each vertical or horizontal segment is labeled based on the normal of its least-square plane. A pair of vertical or horizontal segments is merged if they are neighbors. Through this region growing process, the vertical and horizontal planes are extracted from the range data. The proposed method has a myriad of applications in navigating mobile robots in indoor environments.

This paper presents a glowworm metaphor based distributed algorithm that enables a collection of minimalist mobile robots to split into subgroups, exhibit simultaneous taxis-behavior towards, and rendezvous at multiple radiation sources such as nuclear/hazardous chemical spills and fire-origins in a fire calamity. The algorithm is based on a glowworm swarm optimization (GSO) technique that finds multiple optima of multimodal functions. The algorithm is in the same spirit as the ant-colony optimization (ACO) algorithms, but with several significant differences. The agents in the glowworm algorithm carry a luminescence quantity called luciferin along with them. Agents are thought of as glowworms that emit a light whose intensity is proportional to the associated luciferin. The key feature that is responsible for the working of the algorithm is the use of an adaptive local-decision domain, which we use effectively to detect the multiple source locations of interest. The glowworms have a finite sensor range which defines a hard limit on the local-decision domain used to compute their movements. Extensive simulations validate the feasibility of applying the glowworm algorithm to the problem of multiple source localization. We build four wheeled robots called glowworms to conduct our experiments. We use a preliminary experiment to demonstrate the basic behavioral primitives that enable each glowworm to exhibit taxis behavior towards source locations and later demonstrate a sound localization task using a set of four glowworms