Transformer in Transformer as Backbone for Deep Reinforcement Learning
This work provides an off-the-shelf backbone for deep RL in both online and offline settings, though it is incremental as it builds on existing Transformer combinations.
The paper tackled the challenge of designing pure Transformer-based networks for deep reinforcement learning to avoid the tedious optimization required by mixed-module architectures, and the proposed Transformer in Transformer (TIT) backbone achieved satisfactory performance across different settings.
Designing better deep networks and better reinforcement learning (RL) algorithms are both important for deep RL. This work focuses on the former. Previous methods build the network with several modules like CNN, LSTM and Attention. Recent methods combine the Transformer with these modules for better performance. However, it requires tedious optimization skills to train a network composed of mixed modules, making these methods inconvenient to be used in practice. In this paper, we propose to design \emph{pure Transformer-based networks} for deep RL, aiming at providing off-the-shelf backbones for both the online and offline settings. Specifically, the Transformer in Transformer (TIT) backbone is proposed, which cascades two Transformers in a very natural way: the inner one is used to process a single observation, while the outer one is responsible for processing the observation history; combining both is expected to extract spatial-temporal representations for good decision-making. Experiments show that TIT can achieve satisfactory performance in different settings consistently.