Thilo Grundmann

6D Pose Uncertainty in Robotic Perception

Robotic perception is fundamental to important application areas. In the Joint Research Project DESIRE, we develop a robotic perception system with the aim of perceiving and modeling an unprepared kitchen scenario with many objects. It relies on the fusion of information from weak features from heterogenous sensors in order to classify and localize objects. This requires the representation of wide spread probability distributions of the 6D pose.

Fast parametric viewpoint estimation for active object detection

Most current solutions to active perception planning struggle with complex state representations or fast and efficient sensor parameter selection strategies. The goal is to find new viewpoints or optimize sensor parameters for further measurements in order to classify an object and precisely locate its position. This paper presents an exclusively parametric approach for the state estimation and decision making process to achieve very low computational complexity and short calculation times. The proposed approach assumes a realistic, high dimensional and continuous state space for the representation of objects expressing their rotation, translation and class. Its probability distribution is described by multivariate mixtures of Gaussians which allow the representation of arbitrary object hypotheses. In a statistical framework Bayesian state estimation updates the current state probability distribution based on a scene observation which depends on the sensor parameters. These are selected in a decision process which aims on reducing the uncertainty in the state distribution. Approximations of information theoretic measurements are used as evaluation criteria.

Probabilistic action planning for active scene modeling in continuous high-dimensional domains

In active perception systems for scene recognition the utility of an observation is determined by the information gain in the probability distribution over the state space. The goal is to find a sequence of actions which maximizes the system knowledge at low resource costs. Most current approaches focus either on optimizing the determination of the payoff neglecting the costs or develop sophisticated planning strategies for simple reward models.

Probabilistic Rule Set Joint State Update as approximation to the full joint state estimation applied to multi object scene analysis

One essential capability of service robots lies in the identification and localization of objects in the vicinity of the robot. The extreme computational demands of this high-dimensional state estimation problem require approximations of the joint posterior even for small numbers of objects. A common approach to solve this problem is to marginalize the joint state space and to consider object-related state spaces which are estimated individually under the assumption of statistical independence. In practice, however, this independence assumption is often violated, especially when the objects are located close to each other, which leads to a reduced accuracy of this approximation, compared to the full joint estimation. To address this problem, we propose the new method denoted as Rule Set Joint State Update (RSJSU), which features a better approximation of the joint posterior in the presence of dependencies, and thus leads to better estimation results. We present experimental results in which we simultaneously estimate all six degrees of freedom of multiple objects.

A Gaussian measurement model for local interest point based 6 DOF pose estimation

One of the main challenges for service robots during operation lies in the handling of unavoidable uncertainties which originate from model and sensor inaccuracies and which are characteristic for realistic application scenarios. Robustness under real world conditions can only be achieved when the dominant uncertainties are explicitly represented and purposefully managed by the robots control system. We therefore adopt a probabilistic approach in which perception is regarded as a sequential estimation process and follow a Bayesian filtering methodology. Under these assumptions probabilistic models of the robots perception systems are key. In this paper we shortly describe a model based object recognition and localization system. However, we do not not focus on the 6D pose estimation procedure itself, but on the method to quantify and compute the uncertainty associated with it. We construct a Gaussian approximation of the resulting pose error using the implicit function theorem. It is then used as a proposal density for importance sampling. Our goal is to sample from the measurement model describing 6D object localization based on local features in a Bayesian filtering context.

Local dependency analysis in probabilistic scene estimation

A general solution to the problem of jointly estimating the state of multiple entities is regarded computationally challenging at the time. Most solutions are based on the application of wide assumptions of independence. In many situations and constellations of entities, this is sufficient and leads to high quality results. In some situations as occlusion for instance the assumption of independence is violated heavily resulting in considerable errors. The proposed approach considers local dependencies, allowing to increase the accuracy of the estimation punctually, depending on application requirements, such as high precision localization for grasping operations or rough precision for semantic localization.

Scene Analysis for Service Robots

A scene analysis module for service robots is presented which uses SIFT in a stereo setting, a systematic handling of uncertainties and an active perception component. The system is integrated and evaluated on the DESIRE two-arm mobile robot. Complex everyday scenes composed of various items from a 100-object database are analyzed successfully and efficiently.

Rule Set Based Joint State Update

The accurate localization of the objects in the environment is one of the fundamental preconditions for the reliability of service robots. The majority of algorithms for object localization lacks the ability to integrate physical commonsense knowledge into the recognition process especially, when multiple objects are envolved. Consequently the estimates of such methods often do not comply with basic physical constraints such as that rigid objects should not intersect. In this paper, we present an approach for multi-object localization that is able to consider such physical constraints as statistical dependencies in state estimation processes to increase the localization accuracy. Extensive experiments carried out with a real robot in the context of a service robotics scenario demonstrate the practical usefulness of our approach.

A probabilistic measurement model for local interest point based 6 DOF pose estimation

The ability to recognize objects and to localize them precisely is essential in all service robotic applications. One of the main challenges for service robots during operation lies in the handling of unavoidable uncertainties which originate from model and sensor inaccuracies and are characteristic for realistic application scenarios. Robustness under real world conditions can only be achieved when the dominant uncertainties are explicitly represented and purposefully managed by the robots control system. We therefore adopt a probabilistic approach in which environment perception over time is regarded as a sequential estimation process and follow a Bayesian filtering methodology. Under these assumptions probabilistic models of the robots perception systems play a decisive role. In this paper we describe our object localization system which is based on local features and uses 3D models that are created in an off-line modeling process. A probabilistic model of the errors, which occur in the 6D localization based on local features, is directly derived from the pose reconstruction procedure. Experimental results from an household scenario illustrate the effectiveness of our approach.