Without highly efficient perception, service robotics is not possible: Goal-oriented and safe movement requires having knowledge of the obstacles within the environment and being able to navigate reliably within the room. For objects to be grasped and manipulated, they must first be found, recognised and localised - even under uncooperative conditions where occlusions or unfavourable lighting prevail. Detecting and recognising people as well as reliably identifying gestures are key to being able to understand situations and intentions.

Today, robots function merely in simple, controlled worlds where only few objects are around. For this reason, the aim of DESIRE is to make a significant leap towards making robots suitable for everyday use. In doing so, the core elements involve fast modelling of the obstacle environment using 3D sensor technology, fusion of various sources (sensors, detectors), explicit and consistent treatment of uncertainties, tracking of people over extended periods of time as well as intelligent control of the recognition and analysis algorithms according to a given situation, including "active" mechanisms that use the robot's mobility.

At the same time as algorithms are being developed, a high-performance architecture for robotic perception is emerging. Requirements, metrics and benchmarks are defined in the context of domestic applications; and these form the basis for evaluating the results on the common platform

Scheduled work in the context of perception relates to the safe detection of obstacles, analyzing the scene, as well as perceiving people and situations. For all of the functions, a common sensory platform is used in the project which comprises a colour stereo camera, a 3D camera and a 2D-laser scanner.

• Safe and fast detection and modeling of obstacles in 3D
• Module for the analysis of everyday scenes
• Algorithms for the robust recognition and identification of people and interaction partners
• Robust deictics