J Jos Elfring

Semantic world modeling using probabilistic multiple hypothesis anchoring

In order to successfully perform typical household tasks such as manipulation or navigation, domestic robots need an accurate description of the world they are operating in. Creating and maintaining such a description, in this work referred to as world model, is a non-trivial task in a domestic environment that typically has a high number of objects, and is unstructured and dynamically changing. This work introduces probabilistic multiple hypothesis anchoring to create and maintain a semantically rich world model using probabilistic anchoring. Multiple hypothesis tracking-based data association is included to be able to deal with ambiguous scenarios. Multiple model tracking is included to be able to easily incorporate different kinds of prior knowledge.

The impact of parameter identification methods on drug therapy control in an intensive care unit.

This paper investigates the impact of fast parameter identification methods, which do not require any forward simulations, on model-based glucose control, using retrospective data in the Christchurch Hospital Intensive Care Unit. The integral-based identification method has been previously clinically validated and extensively applied in a number of biomedical applications; and is a crucial element in the presented model-based therapeutics approach. Common non-linear regression and gradient descent approaches are too computationally intense and not suitable for the glucose control applications presented. The main focus in this paper is on better characterizing and understanding the importance of the integral in the formulation and the effect it has on model-based drug therapy control. As a comparison, a potentially more natural derivative formulation which has the same computation speed advantages is investigated, and is shown to go unstable with respect to modelling error which is always present clinically. The integral method remains robust.

Learning intentions for improved human motion prediction

For many tasks, robots need to operate in human populated environments. Human motion prediction therefore is gaining importance. The concept of social forces defines virtual repelling and attracting forces from and to obstacles and points of interest. These social forces can then be used to model typical human movements given an environment and a persons intention. This work shows how such models can exploit typical motion patterns summarized by growing hidden Markov models (GHMMs) that can be learned from data online and without human intervention. An extensive series of experiments shows that exploiting the intended position estimated using a GHMM within a social forces based motion model yields a significant performance gain in comparison with the standard constant velocity-based models.

Active Object Search Exploiting Probabilistic Object–Object Relations

This paper proposes a probabilistic object-object relation based approach for an active object search. An important role of mobile robots will be to perform object-related tasks and active object search strategies deal with the non-trivial task of finding an object in unstructured and dynamically changing environments. This work builds further upon an existing approach exploiting probabilistic object-room relations for selecting the room in which an object is expected to be. Learnt object-object relations allow to search for objects inside a room via a chain of intermediate objects. Simulations have been performed to investigate the effect of the camera quality on path length and failure rate. Furthermore, a comparison is made with a benchmark algorithm based the same prior knowledge but without using a chain of intermediate objects. An experiment shows the potential of the proposed approach on the AMIGO robot.

A robust-control-relevant model validation approach for continuously variable transmission control

High performance continuously variable transmission (CVT) operation requires a reliable control design for its actuation system. The aim of the present paper is to design a high performance robust controller for a range of operating conditions. High performance robust control is achieved by identifying a robust-control-relevant model set that represents the relevant dynamics of the actuation system for a range of operating conditions. Specifically, a new coordinate frame for representing model uncertainty is adopted that transparently connects the size of the model uncertainty and the control criterion, consequently a nonconservative control design can be obtained. Subsequent robust control synthesis reveals that robust performance has been significantly improved over the entire operating range, with respect to both the criterion value and relevant simulated and measured closed-loop step responses.

Multisensor simultaneous vehicle tracking and shape estimation

This work focuses on vehicle automation applications that require both the estimation of kinematic and geometric information of surrounding vehicles, e.g., automated overtaking or merging. Rather then using one sensor that is able to estimate a vehicles geometry from each sensor frame, e.g., a lidar, a multisensor simultaneous vehicle tracking and shape estimation approach is proposed. Advanced measurement models and adequate Bayesian filters enable the shape estimation that is impossible with any of the sensors individually. The use of multiple sensors increases robustness, lowers the complexity of the sensors involved and leads to a gradual loss of performance in case a sensor fails. A series of real world experiments is performed to analyze the performance of the proposed method.

Effective World Modeling: Multisensor Data Fusion Methodology for Automated Driving

The number of perception sensors on automated vehicles increases due to the increasing number of advanced driver assistance system functions and their increasing complexity. Furthermore, fail-safe systems require redundancy, thereby increasing the number of sensors even further. A one-size-fits-all multisensor data fusion architecture is not realistic due to the enormous diversity in vehicles, sensors and applications. As an alternative, this work presents a methodology that can be used to effectively come up with an implementation to build a consistent model of a vehicle’s surroundings. The methodology is accompanied by a software architecture. This combination minimizes the effort required to update the multisensor data fusion system whenever sensors or applications are added or replaced. A series of real-world experiments involving different sensors and algorithms demonstrates the methodology and the software architecture.

Semi-task-dependent and uncertainty-driven world model maintenance

Nearly every task a domestic robot could potentially solve requires a description of the robot’s environment which we call a world model. One problem underexposed in the literature is the maintenance of world models. Rather than on creating a world model, this work focuses on finding a strategy that determines when to update which object in the world model. The decision whether or not to update an object is based on the expected information gain obtained by the update, the action cost of the update and the task the robot performs. The proposed strategy is validated during both simulations and real world experiments. The extended series of simulations is performed to show both the performance gain with respect to a benchmark strategy and the effect of the various parameters. The experiments show the proposed approach on different set-ups and in different environments.