This run only allows the robot to move forwards, turning by having one motor going slower than the other.
Now the evolution cheats and finds a local maxima by simply ignoring the goal and always just turning away from any objects until it they are all out of range. Half way there This just implies that it is more interested in not bumping into things and less interested in the goal.... time to make the goal more appealing, chocolate I think
Wednesday, 24 March 2010
Evolving a diff wheeled robot
I'm evolving a controller for a robot with powered wheels on each side and a non-powered omni-wheel at the back.
The robot needs to get from the start to the goal (shown as a yellow post, in the top left or right of the arena). But it can't bump into any of the cylinders.
1st Try:
The problem with this is that the robot is cheating by going backwards. This is because score is taken away when the front sensors sense close objects.
The robot needs to get from the start to the goal (shown as a yellow post, in the top left or right of the arena). But it can't bump into any of the cylinders.
1st Try:
The problem with this is that the robot is cheating by going backwards. This is because score is taken away when the front sensors sense close objects.
Thursday, 11 March 2010
variable square wave
same again... and i've just spotted a mistake in my graph making code.... so i need to do them all again... bugger
EvoResults-slider2-2d-feetonthefloorsummedscore-tanh-variablesinewavehips-jointsonlyfrontandback-removedsidemotors-bugfix
The hips are on a variable sine wave, varying the frequency, amplitude and y axis offset.
The evolution seems to just push the wave right up high so the motors are always positive and at their limit, sort of vibrating along.
The evolution seems to just push the wave right up high so the motors are always positive and at their limit, sort of vibrating along.
Tuesday, 9 March 2010
EvoResults-slider2-2d-feetonthefloorsummedscore-tanh-sinewavehips-jointsonlyfrontandback-removedsidemotors-bugfix
Fitness: Forward Motion with one foot on the floor
Dimensions of Feedback: 2D
Track: Slider
Generations: 100
Simple ANN using tanh with (1 x sin(0.65 x i)) for the hips
Dimensions of Feedback: 2D
Track: Slider
Generations: 100
Simple ANN using tanh with (1 x sin(0.65 x i)) for the hips
EvoResults-slider2-2d-feetonthefloorsummedscore-tanh-forcedhipswing1sec360degs-jointsonlyfrontandback-removedsidemotors-bugfix
Some nice natural-ish movement on a couple of gens
Monday, 8 March 2010
EvoResults-slider2-2d-feetonthefloorworstscore-tanh-forcedhipswing1sec360degs-jointsonlyfrontandback-removedsidemotors-bugfix
Bugfix run, with some interesting tests
The leg movement on Gen 90 looks pretty nice to me
Saturday, 6 March 2010
Friday, 5 March 2010
Thursday, 4 March 2010
Wednesday, 3 March 2010
Tuesday, 2 March 2010
Friday, 26 February 2010
Thursday, 25 February 2010
Subscribe to:
Posts (Atom)