Version 1 (modified by dhabering, 5 months ago) (diff)


After RoboCup Teammeeting:

We will get pizza, but its complicated.


There was an approach for learning a gripping motion directly for an input image. We may want to try this. Directly get movement commands for given input or just find the workpiece position
Dangerous: changing the gripper needs retrain. Gain: Faster than ICP and if we directly train movement controller we can reduce the mps-align -> plane -> pose pipeline

Pizza is very complicated

Alternative: VISP: CAD Modell + Depth Image + initial position (conveyor plane) to exact position. Native Support for realsense. Continous tracking would enable us to use a closed loop controller

Conveyor Plane: Ransac is quite dangerous and matches any plane it finds. alternative approach: Histogram of normals

In summary: We want to speed up the gripping process. The axis are quit fast, but icp is quite slow. Abortion criteria of ICP are a tuning point
If we can tune ICP to be a lot faster, than it does not make sense to start another approch to get the position. Directly controlling the gripping process may be a valid alternative, though
Picking is more tolerant than putting, it may be feasible to use a faster but less precise alignment method for picking.

Paying pizza is even more complicated

Mario is interested in working on workpiece detection.


Viktor is leaving the FH mid june. Does not know if he is going to come. But wants to. France is nice
David has 60 credits and will be there.
Mario wants but has to check if it fits with his job.
Sebastian yes. Only if women join.
Nic does not know either. Project ends next year but only in december.
Christoph no GO because of BA. France probably. May wants to concentrate on study and may take a break. Tries to get smart factory and gets money to do robocup.
Till is there.
Tarik does not know because he does not know what he wants to do after his master. Maybe doctor at kbsg.
Morian does now know. master thesis.
Daniel has a full time job and does not know.

We need new people.
Agent guys can be found in the kbsg practica. Till asks visual computing institut. Christoph knows 2 interested people. Nick has a student. Other idea: use a specific task as bachelor thesis.

Delivering pizza is also complicated


Bosch is interested but we dont know yet. Hackathon is a lot of work and even more if it should turn out good. We may invite the other teams if we have running field though.

Pizza is there!![[BR]]


Very good baseline. Is robust and a nice framework. Next step is to make more informed decisions about selecting goals, which order to follow etc. Start using planner instead of fixed sequences. Planning in a greater scope (full game) is closer to the actual scenario of the league.

We cant get faster. But the controller is open for improvement.
Multirobot navigation is another point. A bias towards one side could be a solution to situations where two bots are facing each other. Sharing velocity as well to solve close situations.


Be more modular, redesign gripper, maybe remove table. Drive axis even faster (O-Drive, steeper angled spindel). Workpieces may get lost or loose parts while driving due to vibrations.
Put work into robotino transportation.
Robotino 4 is coming (Busen)


Have a proper visualisation of the game state of a robot. Show game time, orders, running skill and goal. Ease communication between operators.