Abstract

This paper proposes a robust approach for image based floor detection and segmentation from sequence of images or video. In contrast to many previous approaches, which uses a priori knowledge of the surroundings, our method uses combination of modified sparse optical flow and planar homography for ground plane detection which is then combined with graph based segmentation for extraction of floor from images. We also propose a probabilistic framework which makes our method adaptive to the changes in the surroundings. We tested our algorithm on several common indoor environment scenarios and were able to extract floor even under challenging circumstances. We obtained extremely satisfactory results in various practical scenarios such as where the floor and non floor areas are of same color, in presence of textured flooring, and where illumination changes are steep.

Abstract

We present a probabilistic method of finding the next best viewpoint that maximizes the chances of finding an object in a known environment for an indoor mobile robot. We make use of the information that is available to a robot in the form of potential locations to search for an object. Extraction of these potential locations and their representation for exploration is explained. This work primarily focuses on placing the robot at its best location in the environment to detect, recognize an object and hence do object search. With experiments done on the exploration, object recognition individually we show the robustness of this approach for object search task. We analyse and compare our method with two other strategies for localizing the object empirically and show unequivocally that the strategy based on the probabilistic formalism in general performs better than the other two.

Abstract

This paper introduces a vision based localization method for large scale urban environments. The method is based upon Bag-of-Words image retrieval techniques and handles problems that arise in urban environments due to repetitive scene structure and the presence of dynamic objects like vehicles. The localization system was experimentally verified it localization experiments along a 5km long path in an urban environment.

Abstract

This paper presents a realtime, incremental multibody visual SLAM system that allows choosing between full 3D reconstruction or simply tracking of the moving objects. Motion reconstruction of dynamic points or objects from a monocular camera is considered very hard due to well known problems of observability. We attempt to solve the problem with a Bearing only Tracking (BOT) and by integrating multiple cues to avoid observability issues. The BOT is accomplished through a particle filter, and by integrating multiple cues from the reconstruction pipeline. With the help of these cues, many real world scenarios which are considered unobservable with a monocular camera is solved to reasonable accuracy. This enables building of a unified dynamic 3D map of scenes involving multiple moving objects. Tracking and reconstruction is preceded by motion segmentation and detection which makes use of efficient geometric constraints to avoid difficult degenerate motions, where objects move in the epipolar plane. Results reported on multiple challenging real world image sequences verify the efficacy of the proposed framework.

Abstract

In this paper we propose a novel physics based motion planning and trajectory generation framework for vehicle operating on uneven terrains. The proposed framework provides for a fully 3D analysis of the dynamic constraints of the vehicle on uneven terrain and hence comes as a better approach than the existing motion planning framework which makes simplifying assumptions for the terrain conditions or the vehicle geometry or both. The entire framework consists of three major parts which are: 1. A framework for determination of the posture of a vehicle in 3D for a given terrain. 2. A framework for determination of maximum feasible velocities and acceleration based on contact and no-slip constraints. 3. Combining the above two framework to generate feasible trajectories for the vehicle. Trajectories are generated through a Dynamic Window paradigm extended to fully 3D terrains, wherein the next best node is selected through a new metric that maximizes the space of feasible velocities and accelerations and reduces the distance to be traversed to the goal.

Abstract

In this paper, we present a monocular camera based terrain classification scheme. The uniqueness of the proposed scheme is that it inherently incorporates spatial smoothness while segmenting a image, without requirement of post-processing smoothing methods. The algorithm is extremely fast because it is build on top of a Random Forest classifier. The baseline algorithm uses color, texture and their combination with classifiers such as SVM and Random Forests. We present comparison across features and classifiers. We further enhance the algorithm through a label transfer method. The efficacy of the proposed solution can be seen as we reach a low error rates on both our dataset and other publicly available datasets.

Abstract

In this paper, we present a practical vision based Simultaneous Localization and Mapping (SLAM) system for a highly dynamic environment. We adopt a multibody Structure from Motion (SfM) approach, which is the generalization of classical SfM to dynamic scenes with multiple rigidly moving objects. The proposed framework of multibody visual SLAM allows choosing between full 3D reconstruction or simply tracking of the moving objects, which adds flexibility to the system, for scenes containing non-rigid objects or objects having insufficient features for reconstruction. The solution demands a motion segmentation framework that can segment feature points belonging to different motions and maintain the segmentation with time. We propose a realtime incremental motion segmentation algorithm for this purpose. The motion segmentation is robust and is capable of segmenting difficult degenerate motions, where the moving objects is followed by a moving camera in the same direction. This robustness is attributed to the use of efficient geometric constraints and a probability framework which propagates the uncertainty in the system. The motion segmentation module is tightly coupled with feature tracking and visual SLAM, by exploring various feed-backs in between these modules. The integrated system can simultaneously perform realtime visual SLAM and tracking of multiple moving objects using only a single monocular camera.

Abstract

The Modular Legged robotic system [1] "Mod-Leg" presented here has been bio-inspired from a Snake's vertebrae and a caterpillar's legged structure. The system can be configured to a 4-legged robotic dog, a hexapod, a caterpillar and a Snake robot. This robot's novel design achieves compliance to the terrain using a combination of legs and electronically actuated universal spine. A unique simulator has been designed for this purpose. Some of the things we learned while developing this robotic system have been presented below.

Abstract

The process of scancorrelation is very useful in map building and localization in environments without any easily discernible landmarks. It involves estimating the relative transform between two given scans of data. This can be formulated as a posterior estimation problem over the transformation space. Since the posterior, the action model and the likelihood function are mostly nonlinear, a particle filter is used to estimate this posterior [1]. A GMM based correlation scheme has been used in [2] to drive the scan correlation process. In this paper, we propose a 3 step particle filtering approach that combines the simplicity and efficiency of the particle filter with the robustness of the GMM based correlation scheme. Experimental results are provided to verify the effectiveness of this approach.

Abstract

The task of the robot in localization is to find out where it is, through sensing and motion. In environments which possess relatively few features that enable a robot to unambiguously determine its location, global localization algorithms can result in ‘multiple hypotheses’ locations of a robot. This is inevitable with global localization algorithms, as the local environment seen by a robot repeats at other parts of the map. Thus, for effective localization, the robot has to be actively guided to those locations where there is a maximum chance of eliminating most of the ambiguous states — which is often referred to as ‘active localization’. When extended to multi-robotic scenarios where all robots possess more than one hypothesis of their position, there is an opportunity to do better by using robots, apart from obstacles, as ‘hypotheses resolving agents’. The paper presents a unified framework which accounts for the map structure as well as measurement amongst robots, while guiding a set of robots to locations where they can singularize to a unique state. The strategy shepherds the robots to places where the probability of obtaining a unique hypothesis for a set of multiple robots is a maximum. Another aspect of framework demonstrates the idea of dispatching localized robots to locations where they can assist a maximum of the remaining unlocalized robots to overcome their ambiguity, named as ‘coordinated localization’. The appropriateness of our approach is demonstrated empirically in both simulation & real-time (on Amigo-bots) and its efficacy verified. Extensive comparative analysis portrays the advantage of the current method over others that do not perform active localization in a multi-robotic sense. It also portrays the performance gain by considering map structure and robot placement to actively localize over methods that consider only one of them or neither. Theoretical backing stems from the proven completeness of the method for a large category of diverse environments.