Felix von Hundelshausen

FOCUS: a generalized method for object discovery for robots that observe and interact with humans

The essence of the signal-to-symbol problem consists of associating a symbolic description of an object (e.g., a chair) to a signal (e.g., an image) that captures the real object. Robots that interact with humans in natural environments must be able to solve this problem correctly and robustly. However, the problem of providing complete object models a priori to a robot so that it can understand its environment from any viewpoint is extremely difficult to solve. Additionally, many objects have different uses which in turn can cause ambiguities when a robot attempts to reason about the activities of a human and their interactions with those objects. In this paper, we build upon the fact that robots that co-exist with humans should have the ability of observing humans using the different objects and learn the corresponding object definitions. We contribute an object recognition algorithm, FOCUS, that is robust to the variations of signals, combines structure and function of an object, and generalizes to multiple similar objects. FOCUS, which stands for Finding Object Classification through Use and Structure, combines an activity recognizer capable of capturing how an object is used with a traditional visual structure processor. FOCUS learns structural properties (visual features) of objects by knowing first the objects affordance properties and observing humans interacting with that object with known activities. The strength of the method relies on the fact that we can define multiple aspects of an object model, i.e., structure and use, that are individually robust but insufficient to define the object, but can do when combined.

An Omnidirectional Vision System That Finds and Tracks Color Edges and Blobs

We describe the omnidirectional local vision system developed for the FU-Fighters, a RoboCup F180 league soccer team. A small video camera mounted vertically on top of the robots looks at a concave parabolic mirror placed above the camera that reflects the field around. The image is sent via a radio link to an external PC for processing. Our computer vision system can find the ball and detect other robots as obstacles. The walls of the field are also recognized and are used to determine the initial position of the robot. In order to be able to process the video stream at full frame rate the movement of all objects is tracked, including the walls of the field. The key idea of our approach is to predict the location of color edges in the next frame and to search for such color transitions along lines that are perpendicular to the edge.

3D outline contours of vehicles in 3D-LIDAR-measurements for tracking extended targets

Tracking of extended targets in high definition 360 degree 3D-LIDAR (Light Detection and Ranging) measurements is a challenging task. It is a key component in robotic applications and is relevant to collision avoidance and autonomous driving. This paper presents a robust method to determine the 3D outline contour of vehicles in disordered 3D-LIDAR measurements while using several geometrical vehicle-specific constraints. In addition, the 3D outline contour contains information on the local reliability of the contour. A weighted registration approach allows calculating the velocity of consecutive 3D outline contours directly. The approach is tested with real sensor data. A robot car equipped with an inertial measurement unit serves as ground truth.

Geometric-Model-Free Tracking of Extended Targets Using 3D-LIDAR-Measurements

Tracking of extended targets in high definition, 360-degree 3D-LIDAR (Light Detection and Ranging) measurements is a challenging task and a current research topic. It is a key component in robotic applications, and is relevant to path planning and collision avoidance. This paper proposes a new method without a geometric model to simultaneously track and accumulate 3D-LIDAR measurements of an object. The method itself is based on a particle filter and uses an object-related local 3D grid for each object. No geometric object hypothesis is needed. Accumulation allows coping with occlusions. The prediction step of the particle filter is governed by a motion model consisting of a deterministic and a probabilistic part. Since this paper is focused on tracking ground vehicles, a bicycle model is used for the deterministic part. The probabilistic part depends on the current state of each particle. A function for calculating the current probability density function for state transition is developed. It is derived in detail and based on a database consisting of vehicle dynamics measurements over several hundreds of kilometers. The adaptive probability density function narrows down the gating area for measurement data association. The second part of the proposed method addresses weighting the particles with a cost function. Different 3D-griddependent cost functions are presented and evaluated. Evaluations with real 3D-LIDAR measurements show the performance of the proposed method. The results are also compared to ground truth data.

A constructive feature detection approach for robotic vision

We describe a new method for detecting features on a marked RoboCup field. We implemented the framework for robots with omnidirectional vision, but the method can be easily adapted to other systems. The focus is on the recognition of the center circle and four different corners occurring in the penalty area. Our constructive approach differs from previous methods, in that we aim to detect a whole palette of different features, hierarchically ordered and possibly containing each other. High-level features, such as the center circle or the corners, are constructed from low-level features such as arcs and lines. The feature detection process starts with low-level features and iteratively constructs higher features. In RoboCup the method is valuable for robot self-localization; in other fields of application the method is useful for object recognition using shape information.

Active Monte Carlo recognition

In this paper we introduce Active Monte Carlo Recognition (AMCR), a new approach for object recognition. The method is based on seeding and propagating relational particles that represent hypothetical relations between low-level perception and high-level object knowledge. AMCR acts as a filter with each individual step verifying fragments of different objects, and with the sequence of resulting steps producing the overall recognition. In addition to the object label, AMCR also yields the point correspondences between the input object and the stored object. AMCR does not assume a given segmentation of the input object. It effectively handles object transformations in scale, translation, rotation, affine and non-affine distortion. We describe the general AMCR in detail, introduce a particular implementation, and present illustrative empirical results.

Determining the outline contour of vehicles in 3D-LIDAR-measurements

This paper presents a novel and robust method to determine the outline contour of vehicles in 3D-LIDAR (Light Detection and Ranging) measurements. To calculate the outline contour, a vehicle is described by its geometrical properties. These properties are used as constraints to fit a surface to unordered, scattered and error-contaminated 3D measurements. The surface can be used to calculate a corresponding 2D outline contour. The algorithm is tested with two different laser scanners. One scanner has 64, the other only 4 layers.

FU-Fighters Omni 2001 (Local Vision)

Currently, the Small Size League is the only RoboCup competition that permits the use of external sensing systems. Most teams use a color camera that is mounted above the field to determine the positions of the robots and the ball. Since this simplified setup is not compatible with the idea of autonomous robots, we decided to build a second F180 team for the World Championships 2001 in Seattle, consisting of (only) three robots using local omnidirectional vision, the “FU-Fighters Omni”.

Driving with tentacles - Integral structures for sensing and motion

In this paper we describe a LIDAR-based navigation approach applied at both the C-Elrob (European Land Robot Trial) 2007 and the 2007 DARPA Urban Challenge. At the C-Elrob 2007 the approach was used without any prior knowledge about the terrain and without global positioning system (GPS). At the Urban Challenge the approach was combined with a GPS-based path follower. At the core of the method is a set of “tentacles” that represent precalculated trajectories defined in the ego-centered coordinate space of the vehicle. Similar to an insect’s antennae or feelers, they fan out with different curvatures discretizing the basic driving options of the vehicle. We detail how the approach can be used for exploration of unknown environments and how it can be extended to combined GPS path following and obstacle avoidance allowing safe road following in case of GPS offsets. C