The goal in sensor fusion is to utilize information from spatially separated sensors of the same kind (so called sensor networks), sensors of different kind and finally on a more abstract level information sources in general in terms as for example geographical information systems. The current activities concern

Particle filter theory The particle offers a general algorithm for state estimation in the models above, with potentially better peformance than the classical extended Kalman filter (EKF) for nonlinear or non-Gaussian systems. Our contributions involve:

Convergence analysis for the state estimate, where results with relaxed asssumptions have been presented.

Marginalization to mitigate the complexity of the particle filter, which in the end allows a scalable algorithm.

Applications

Fusion of sensor observations and GIS. Concrete applications are terrain navigation of aircraft using altitude GIS, terrain navigation of underwater vehicles using bottom depth GIS, localization in road networks using road GIS, surface ship navigation using radar and sea chart GIS, etc.

SLAM Simultaneous Localization And Mapping (SLAM) aims at solving to tasks in the same filter, namely localization of the host vehicle and at the same time building a map (GIS) of the surrounding. Both EKF and PF based approaches have been proposed, and we are actively developing these approaches for different applications (unmanned aerial vehicles for instace).

Collision Avoidance (CA) is one important application of state estimation, where the estimated state is used to assess the risk for a conflict, and also for evaluating different evasive maneuvers.

Automotive CA is researched with Volvo Car in a serie of joint projects.

Airborne CA, often referred to as sense and avoid, is a joint project with SAAB Aerospace.