|
Sony AIBO MoviesHere are links to our most recent Sony AIBO movies. Click on the speed links below to jump to what your interested in. For more information about the technical details behind these movies, please visit our publications and/or Sony AIBO legged league pages: AcceleratorsMoviesThese vides were taken from RoboCup US Open 2004, held in New Orleans. Our team CMPack'04 won first place.
These vides were taken from RoboCup 2004, held in Lisbon, Portugal.
2003RoboCup-CMPack03_v_Griffith 2003RoboCup-CMPack03_v_Germany_Quarterfinal
These are the videos from the RoboCup American Open'03, held in Pittsburgh at Carnegie Mellon University during May of 2003. These are the videos from RoboCup 2002 held in Fukuoka, Japan, during June of 2002. Our team came in first place.
These videos show our general obstacle avoidance algorithm, called Visual Sonar. Essentially, objects that do not match the color labelling of the ground are projected onto a local map, which is used for obstacle avoidance. See our publications for further details. These two videos show the robot-eye view of the world. The robot is trying to kick the ball into the goal. The output after CMVision color segmentation is logged and reproduced as the video you see. It runs at the full frame rate of 25Hz. There are many sources of motion errors. These videos show examples due to falls, being picked up, hooked on another robot, or pushed by another robot. For high performance, motion parameters must be tuned to the environment. If the environment changes, these parameters can suddenly perform poorly. Here we show a Sony AIBO kicking with parameters tuned to the lab, and RoboCup. The difference in performance is remarkable.
Sony AIBO's are remarkably versatile. Here we have a robot climbing into the field. The wall is 10cm tall, and is a signifcant fraction of the robot height. Environmental_State_Detection_and_Recognition Detecting changed environmental and/or sensor states is a challenging problem, and one that is extremely relevant to robot control problems. We are investigating non-parametric, statistical techniques to segment and recognize different sensory states in an on-line, unsupervised fashion. These videos show the need for the technique. One video shows the robot kicking the ball with color thresholds tuned to the bright lab. The second video shows the same parameters when the lights are dimmed. Our new technique allows us to overcome this limitation by recognizing the different environmental condition and switching the color thresholds to the appropriate set for the conditions. Please see our publications for more details.
File generated on Tue Jul 19 00:00:01 2005 |
|||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
|