CoTeSys-ROS Fall School Munich November 3 th 2010 General Concept & First Demonstrator Jos Elfring Rob Janssen
Dec 23, 2015
CoTeSys-ROS Fall SchoolMunich November 3th 2010
General Concept & First Demonstrator
Jos ElfringRob Janssen
General concept• Humans in day2day tasks
• 90% feedforward – learning and memorization• 10% feedback – sensory feedback
[email protected]@tue.nl
General concept• Humans in day2day tasks
• 90% feedforward – learning and memorization• 10% feedback – sensory feedback
• RoboEarth – learning and memorization• Share any reusable knowledge between robots• World-Wide-Web for and by robots
[email protected]@tue.nl
General concept• Humans in day2day tasks
• 90% feedforward – learning and memorization• 10% feedback – sensory feedback
• RoboEarth – learning and memorization• Share any reusable knowledge between robots• World-Wide-Web for and by robots
• Four year EC funding (FP7)
[email protected]@tue.nl
Concept
[email protected]@tue.nl
Concept
[email protected]@tue.nl
Concept
[email protected]@tue.nl
Concept
[email protected]@tue.nl
Example
[email protected]@tue.nl
Example
[email protected]@tue.nl
Bed
Cabinet
Power plug
Bed
Cabinet
Power plug
Bed
Cabinet
Power plug
‘Birth of RoboEarth workshop’Two classes of robots sharing knowledge through RoboEarth
TUE Eindhoven bot RFC Stuttgart bot
[email protected]@tue.nl
Experiment description
• The two classes will collaboratively explore a maze• Sharing of knowledge on two levels
• Sharing of learned inputs for moving 1[m] front, back, left and right within class
• Sharing of environmental knowledge on the maze between classes
[email protected]@tue.nl
Visualisation
1. Learning input primitives (using Iterative learning Control)
[email protected]@tue.nl
1 [m]
Visualisation
1. Learning input primitives (using Iterative learning Control)
[email protected]@tue.nl
1 [m]
1 [m]
Visualisation
1. Learning input primitives (using Iterative learning Control)
[email protected]@tue.nl
1 [m]
1 [m]
1 [m]
Visualisation
1. Learning input primitives (using Iterative learning Control)
[email protected]@tue.nl
1 [m]
1 [m]
1 [m]
1 [m]
Visualisation
1. Learning input primitives (using Iterative learning Control)
[email protected]@tue.nl
1 [m]
1 [m]
1 [m]
1 [m]
Visualisation
1. Learning input primitives (using Iterative learning Control)
[email protected]@tue.nl
1 [m]
1 [m]
1 [m]
1 [m]
Visualisation
2. Use primitives to collaboratively explore a maze (using Q-learning)
[email protected]@tue.nl
Summary
All though imposed methods in this experiment rather trivial, it can been shown that
• robots benefit from sharing knowledge (increased efficiency)• scalable• many forms of knowledge can be shared (object information,
environmental data, action recipes)• RoboEarth databases function as backup servers• sharing through WWW can be used globally
[email protected]@tue.nl
[email protected]@tue.nl