Self-supervised Relational RL with Independently Controllable Subgoals
The dataset used in the paper is a multi-object environment with a robotic arm and multiple objects to manipulate. The agent learns to control the objects independently and solve a compositional goal.
BibTex:
