Robot Uses Imagination To Learn How To Place New Objects - Cornell

Personal robots should have the ability to place objects in a proper way at a preferred location. Our homes are having a large variety of object types as well as of placing areas. So, the algorithms guiding the robots should be able to manipulate the decisions before placing the objects. And Ashutosh Saxena's Lab at #-Link-Snipped-# is trying to achieve just that. They are making the robots learn to use their imaginations to figure out the best possible location to place an object. So, if the robot finds a TV remote lying on the floor, it should place it on a table near the couch instead of the place next to the TV, because thats where a human would want to keep it.



So, the Cornell team put up virtual humans in a 3D scenes and let the robot imagine the movements and analyze those scenes. The robot then uses a supervised learning algorithm to finding a good placement for an object in the giving placing area. It starts learning to combine the features that capture support, stability and preferred placements using a shared sparsity structure in the parameters. If the robot has never seen the object or the placing area ever before it can imagine and make good predictions. See a video -



Via: #-Link-Snipped-#