Domain_Randomization_for_Transferring_Deep_Neural_Networks_from_Simulation_to_the_Real_World
-
tests the feasibility of domain randomization in bridging the “reality gap”
-
focus on image-based tasks and proposes to randomize the rendering of images in simulation
-
test its approach on two tasks, object-localization and using the pre-trained object-localization network as a feature extractor for an off-the-shelf motion planning package to perform object picking
-
argues that testing on object-localization makes sense because it serves as stepping stone to more general robotic manipulation skill
Sim_to_Real_Transfer_of_Robotic_Control_with_Dynamics_Randomization
-
propose to bridge the “reality gap” by randomizing the dynamics of the simulator during training
-
demonstrated on an object-pushing task using a robotic arm