Who’s the boss?
Robogeek doesn’t know what he has constructed here. Clearly the object recognition is state-of-the art, but adding the object-identity module and make it self-learn didn’t give the results he expected. Maybe Igor’s reinforcement techniques (his foot) had actually influenced the weightings. RG needs a plan here!
April 26th, 2013 at 11:18 am
Looking at our robots I would be happy to get this result as long as there is a different response to pictures of house, bridge and cardboard box
Also reminds me of this comic: http://abstrusegoose.com/496
You can improve on the results by maybe changing the test set to only one positive