With only four weeks to go before Tech United will leave for Istanbul, pressure starts to rise. This has its effect on the intensity of team evenings but luckily also on the progress that’s been made! Below you’ll read about our most important breakthroughs.
Crucial for the TURTLEs is 3D ball recognition, which is very important in order to improve the response of the goalkeeper and to prevent the robots from losing the ball once it bounces instead of rolls over the field. Three different tracks that eventually lead to reliable 3D ball recognition are being explored.
The past couple of days, each of these tracks experienced its individual breakthrough. The laser scanner mounted high on the goalkeeper detected lob balls, the ball recognition via the frontcam was demonstrated and the Xbox Kinect also produced a useful ball detection! Still much needs to be done in order to properly tune this new hardware with the TURTLE system but when the team is able to keep working in this pace it will definitely be finished before Istanbul.
Another breakthrough occurred with intercepting the ball. The robots now are better able to think ahead, allowing them to better anticipate on fast rolling balls. This skill is really important while passing the ball and will eventually speed up the game of soccer.
AMIGO learns to forget
Also our RoboCup@Home department made impressive progress. The worlmodel, constructed by the robot to map its surroundings, has been improved. At first AMIGO stored the exact position of every object it saw in a certain room. All this logging of data often caused the model to contain way too many objects, sometimes it even showed objects that were gone already. Therefore the robot now actually sees a difference between a table (whose position should be remembered long time) and a human (whose position can be forgotten after a certain time).Next to the improved speed in constructing the worldmodel, also face recognition has been speeded up significantly. Because the software now is able to determine the stance of a body, it can draw a small square around the face of a person it is looking at. Face recognition applied to this small image is of course much faster than using the complete image.
'