Learning State Representations from Random Deep Action-conditional Predictions
This work offers a surprisingly simple and effective method for learning state representations for reinforcement learning agents, potentially simplifying the design of auxiliary tasks for RL researchers.
This paper empirically demonstrates that random deep action-conditional predictions, used as auxiliary tasks, can learn effective state representations for reinforcement learning. These representations achieve control performance competitive with state-of-the-art hand-crafted auxiliary tasks in Atari and DeepMind Lab environments.
Our main contribution in this work is an empirical finding that random General Value Functions (GVFs), i.e., deep action-conditional predictions -- random both in what feature of observations they predict as well as in the sequence of actions the predictions are conditioned upon -- form good auxiliary tasks for reinforcement learning (RL) problems. In particular, we show that random deep action-conditional predictions when used as auxiliary tasks yield state representations that produce control performance competitive with state-of-the-art hand-crafted auxiliary tasks like value prediction, pixel control, and CURL in both Atari and DeepMind Lab tasks. In another set of experiments we stop the gradients from the RL part of the network to the state representation learning part of the network and show, perhaps surprisingly, that the auxiliary tasks alone are sufficient to learn state representations good enough to outperform an end-to-end trained actor-critic baseline. We opensourced our code at https://github.com/Hwhitetooth/random_gvfs.