Björn Jensen

Robox at Expo.02: A large-scale installation of personal robots

In this paper we present Robox, a mobile robot designed for operation in a mass exhibition and the experience we made with its installation at the Swiss National Exhibition Expo.02. Robox is a fully autonomous mobile platform with unique multi-modal interaction capabilities, a novel approach to global localization using multiple Gaussian hypotheses, and a powerful obstacle avoidance. Eleven Robox ran for 12 hours daily from May 15 to October 20, 2002, traveling more than 3315 km and interacting with 686,000 visitors.

Multisensor on-the-fly localization: Precision and reliability for applications

This paper presents an approach for localization using geometric features from a 360 laser range finder and a monocular vision system. Its practicability under conditions of continuous localization during motion in real time (referred to as on-the-fly localization) is investigated in large-scale experiments. The features are infinite horizontal lines for the laser and vertical lines for the camera. They are extracted using physically well-grounded models for all sensors and passed to a Kalman filter for fusion and position estimation. Positioning accuracy close to subcentimeter has been achieved with an environment model requiring 30 bytes/m 2 . Already with a moderate number of matched features, the vision information was found to further increase this precision, particularly in the orientation. The results were obtained with a fully self-contained system where extensive tests with an overall length of more than 6.4 km and 150,000 localization cycles have been conducted. The final testbed for this localization system was the Computer 2000 event, an annual computer tradeshow in Lausanne, Switzerland, where during 4 days visitors could give high-level navigation commands to the robot via a web interface. This gave us the opportunity to obtain results on long-term reliability and verify the practicability of the approach under application-like conditions. Furthermore, general aspects and limitations of multisensor on-the-fly localization are discussed.

Deriving and matching image fingerprint sequences for mobile robot localization

Proposes a method for creating unique identifiers, called fingerprint sequences, for visually distinct locations by recovering statistically significant features in panoramic color images. Fingerprint sequences are expressive enough for mobile robot localization, as demonstrated using a minimum energy sequence-matching algorithm that is described. Empirical results in two different places demonstrate the reliability of the system for global localization on a Nomad Scout mobile robot.

The interactive autonomous mobile system RoboX

In this paper we present an autonomous mobile robot system allowing for complex collaborative interaction with non-experienced users while giving a preprogrammed tour of a public exposition. Several modalities of man-machine communication are used simultaneously. To ensure secure and reliable operation we present a layered development approach centered on the programming system SOUL. At the Swiss National Exhibition Expo.02, ten RoboX systems will interact with hundreds of visitors per day, during the 5-month period, seven days a week, ten hours per day. First tests in a real world environment and an analysis of its interaction with visitors enable us to judge the credibility of the tour-guide robot and the reliability of its operation for this outstanding event.

Scan alignment with probabilistic distance metric

Scan alignment estimates the relative robot position from corresponding sets of data by identifying the transformation that minimizes a distance metric on these sets. Here, we present a method (SLIP) establishing correspondences between points based on a novel probabilistic distance metric to allow robust detection of outliers. This metric takes into account sensor noise and robot position uncertainty. Outliers are detected as elements with none but low probability links among all correspondences. To achieve scan alignment an inverse model is applied on the links, estimating robot position and reducing position uncertainty. Results of SLIP preserving all links and a computationally more efficient variant retaining only the most probable link are compared to standard ICP for tests with scan data and artificially inserted outliers. Additionally SLIP was used to built maps of an office environment from scan series. It was found to correct position errors and reject outliers from artificial data and real scans successfully.

Towards Real-Time Sensor-Based Path Planning in Highly Dynamic Environments

This paper presents work on sensor-based motion planning in initially unknown dynamic environments. Motion detection and probabilistic motion modeling are combined with a smooth navigation function to perform on-line path planning and replanning in cluttered dynamic environments such as public exhibitions. The SLIP algorithm, an extension of Iterative Closest Point, combines motion detection from a mobile platform with position estimation. This information is then processed using probabilistic motion prediction to yield a co-occurrence risk that unifies dynamic and static elements. The risk is translated into traversal costs for an E* path planner. It produces smooth paths that trade off collision risk versus detours.

Toward Online Probabilistic Path Replanning in Dynamic Environments

This paper presents work on sensor-based motion planning in initially unknown dynamic environments. Motion detection and modeling are combined with a smooth navigation function to perform on-line path planning in cluttered dynamic environments. The SLIP algorithm, an extension of iterative closest point, combines motion detection from a mobile platform with position estimation. This information is used via probabilistic prediction to estimate a traversal risk function that unifies dynamic and static obstacles. The risk is fed to E* and leads to smooth paths that trade off collision risk versus detours

Narrative-level visual interpretation of human motion for human-robot interaction

Compelling human-robot interaction demands high level perception of human behavior by the robot. In this paper we describe a visual perception system that provides high-level, narrative interpretation of human behavior in relation to the robot. The vision system has been implemented using Firewire digital camera technology and has been tested in public venues at The Robotics Institute.

Using EM to detect motion with mobile robots

In this paper we present a new method to detect motion from mobile platforms using laser range data. Motion can be found as differences in successive scans. The main challenge in doing so from a mobile platform is to distinguish differences originating from the platforms own motion from those caused by objects moving in the robots vicinity. We tackle this combining localization and data association based on an a-priori obtained map. Localization and data association is done using the EM-algorithm. Elements, which are not in the map, are singled out as outliers. Subtracting them over time provides motion information. To reduce the complexity of each iteration step we chose a feature-based environment model, which reduces the computation required to a fraction. We use simulations to test our method against a known ground-truth. Results based on real-world data from the exhibition Robotics@Expo.02 are used to evaluate the proposed method under real-world conditions in highly dynamic situations with several hundred visitors per hour.

2D laser-based probabilistic motion tracking in urban-like environments

All over the world traffic injuries and fatality rates are increasing every year. The combination of negligent and imprudent drivers, adverse road and weather conditions produces tragic results with dramatic loss of life. In this scenario, the use of mobile robotics technology onboard vehicles could reduce casualties. Obstacle motion tracking is an essential ability for car-like mobile robots. However, this task is not trivial in urban environments where a great quantity and variety of obstacles may induce the vehicle to take erroneous decisions. Unfortunately, obstacles close to its sensors frequently cause blind zones behind them where other obstacles could be hidden. In this situation, the robot may lose vital information about these obstructed obstacles that can provoke collisions. In order to overcome this problem, an obstacle motion tracking module based only on 2D laser scan data was developed. Its main parts consist of obstacle detection, obstacle classification, and obstacle tracking algorithms. A motion detection module using scan matching was developed aiming to improve the data quality for navigation purposes; a probabilistic grid representation of the environment was also implemented. The research was initially conducted using a MatLab simulator that reproduces a simple 2D urban-like environment. Then the algorithms were validated using data samplings in real urban environments. On average, the results proved the usefulness of considering obstacle paths and velocities while navigating at reasonable computational costs. This, undoubtedly, will allow future controllers to obtain a better performance in highly dynamic environments.