Benjamin Ellenberger

Create a track where you can choose the creature that needs to learn, the controller it runs with (its original controller, a neural controller etc.) Fitness function, environment etc. Freeze...

Caffe research and implementation

5

Different libraries: - http://openann.github.io/OpenANN-apidoc/OtherLibs.html - http://leenissen.dk/fann/wp/ Get more input from Danny and GVGAI.

The third person camera system: - http://www.ogre3d.org/tikiwiki/tiki-index.php?page=3rd+person+camera+system+tutorial&structure=Tutorials

Make it easy to check out and compile

1

This task is optional and might be impossible to do for cross-plattform.

In recurrent networks there are no constraints on the topology of the neural graph and connections can form feed- back loops.Signal propagation works by evaluating neurons in random order,

Final positions of all robots discovered in selected typical runs is shown for the swimming experiment (8a, 8b) and for the barrier avoidance experiment (8c, 8d). Novelty search in both...

Simulation experiences: - http://repo.myorobotics.eu/docs/caliper/simulator.html - http://roy-t.nl/index.php/2013/06/29/a-self-balancing-creature/ - http://www.physicsbasedanimation.com/2014/11/16/realistic-biomechanical-simulation-and-control-of-human-swimming/ - http://animatlab.com/Help/Documentation/Biomechanical-Editor/Bullet-Physics-Engine - http://www.rocket5studios.com/tutorials/approaches-to-animating-quadrupeds-the-walk-cycles/ Ogre-ragdoll physics(?) - http://www.ogre3d.org/tikiwiki/tiki-index.php?page=Ragdolls Neural networks library - https://github.com/antinucleon/cxxnet

Give the option to use muscle driven (simple limited) or servo driven (rather non-limited) actuation. Let the EA evaluate what is good for what (different terrains, gravity, friction etc.)