=== ActuatorControl === * UTS walk and kicks * We need to turn faster. Look at UPenn and Austia! Yes, UPenn has a very fast diagonal walk. It seemed even faster forward walk --> learn short steps walk (small PG) and learn * A specialised short-step walk for close in to the ball. Quicker steps, faster reaction times * Omnidirectional walk. === Vision === * German's visual obstacle detection. * Use maybe colors in sanity checks better. Can sanity checks be learned (incrementally)? * Incremental colour classification. How the dog improve vision while running. I can't find any papers on this topic. * Optical Flow in colour classification. atm, the classified picture is static, is it different if picture when the camare is moving is classified as well. optical flow might help : http://robotics.eecs.berkeley.edu/~sastry/ee20/vision3/node2.html We can record a small stream of video from the camera, generate optical flow. The first frame is classified and the optical flow can be used to automatically classify the rest. This might be better if the color classification can be learnt incrementally. === GPS === * Can GPS help sanity checks? or vice versa, can "GPS sanity" help localisation, for example, the dog GPS tells it is in front of the goal, without any other obstacle between the goal and itself, but vision cannot see any goal. is there any thing wrong with it? * How to calibrate GPS? in other words, calibrate objects' mean and variance. (The current system seems to put quite arbitrary values). How to engineer GPS? * Use the overhead camera? Don't waste the cameras. Not enough people to work on the overhead this year. === Behaviours === * Passing? * After seeing what the ERS7 can do in this competition, how the UTS dogs handle ball, how fast the dog walks, how German localisation performs, I believe so quite strongly. It will happen at some stage, because it really depends on other modules --- Kim * How do people (learn to) kick/catch a ball? We don't compute the differential equations - we learn patterns and responses. Try pattern recognition (neural nets) for complex skills? * Any similar task would be: put a number of standing dogs on the field, can you make your dog dribble the ball though, and score the goal. (The UTS has thought about this in their open challenge, but they kind of cheating to have the dog hold the ball all the way. By dribbling I mean grabbing, releasing, turning...as long as the ball is not held) * UTS (or better) ball grabbing (must, very important) === Misc: Ideas about ideas === * How can we beat 40 PhD (there are only 7 PhD students in the GermanTeam) in the current champion team. Do we need ideas? How many? How great the idea must be? Which ideas would work? which ideas would not? How can make those ideas happen? Well this is understandable, out of the 40 people in the team only 7 were PHDs, the other 33 were professors and lecturers, something like that. * Sometimes, many small ideas is better than one great idea.