#robotics Logs

Dec 13 2018

09:08 AM rue_mohr: any heard of hasel actuators?
09:14 AM veverak: sounds cool
09:18 AM rue_mohr: hmm electrostatic...
09:22 AM veverak: HA
09:22 AM veverak: neuron network is on 95%
09:22 AM rue_mohr: ?
09:24 AM veverak: had to train neuronw network as assigment at university
09:24 AM veverak: pictures of numbers recognition
09:24 AM veverak: 95% success rate on 10k dataset
09:26 AM rue_mohr: sounds right
09:27 AM rue_mohr: thresholded outputs or real values
09:28 AM rue_mohr: "kinda looks like a 5, more looks like a 6"?
09:28 AM veverak: output layer are 10 neurons, the biggest one is selected
09:28 AM rue_mohr: select = threshold
09:28 AM veverak: yup
09:29 AM rue_mohr: there is a lot of value knowing that it still 45% looksl ike a 5
09:29 AM rue_mohr: bbl work
09:40 AM veverak: potentially yeah
09:40 AM veverak: but tbh I just wanted to solve the assigment and move on
09:45 AM veverak: https://imgur.com/a/jLsaLSH
09:45 AM veverak: well
09:45 AM veverak: the path was planned
09:45 AM veverak: but it sucks
09:45 AM veverak: too close to the columns
09:45 AM veverak: I need better condition for collision avoidance
09:46 AM veverak: 'body not in colision with environet && all legs workareas in collision with environment' is not good enough evidently
09:47 AM veverak: maybe select 'folded' leg configuration and use that?
09:47 AM veverak: 'in the pose, folded leg config can't be in collision with environment'
07:46 PM rue_mohr: well did you use tensor flow for the NN character recg?
11:42 PM rue__ is now known as rue_mohr