Many people have contributed to the work shown on this page. They include Chris Atkeson, Gordon Cheng, Chris Gaskett (combining foveal and peripheral vision), Marcia Riley (Okinawan dancing) , Darrin Bentivegna (air-hockey), Tomohiro Shibata (smooth pursuit), and others.

We use peripheral vision to detect and pursue objects of interest based on shape and color models. A detection event triggers the robot to direct its eyes towards the object, thus making a more detailed analysis of the observed objects in higher resolution foveal images feasible. The recognition is based on principal component analysis and is performed while the robot actively pursues the detected object. Once the desired object is recognized, the robot reaches for it while ignoring other objects.

This video shows the estimation of human walking from video using an ellipsoidal body model. The applied estimation method is based on a robust optimization approach. The observed motion is transferred to a computer graphics character and to our humanoid robot DB.

We use a marker-based tracking system Optotrak to reproduce more complex motions. Here you can see the reproduction of Okinawan dance movements.

We developed a system for real-time visual detection and tracking (up to 60 Hz until now) using shape and color information. We made use of probabilistic techniques and affine warping to achieve efficient and reliable operation of the system. This video shows real-time mimicking of human hand motion based on visual information provided by the robot's eyes. The system detects and tracks human body parts such as faces and hands and estimates their 3-D motion using Kalman filter and random jerk motion model in real-time.

The same visual system was applied to keep human faces in the foveal region of the robot's left eye. The motor control subsystem estimates the nonlinear dynamics of a 2-D visual target and of an oculomotor system in real-time.

Our humanoid robot DB can also play air hockey. To successfully play the game, the robot needs to keep track of 6 or more blobs at a relatively high frame rate (60 Hz). Here you can see the results of tracking 9 blobs. Various learning algorithms were applied to improve the air-hockey abilities of DB.

We also experimented with a high speed camera to capture motions at higher frame rates. This movie presents the estimation of human walking at 250 Hz. All captured frames are displayed in the first part of a video, while the second part shows motion at normal speed (30 fps).