Implicit Two-Tower Policies
This work addresses computational efficiency and performance in reinforcement learning for domains like robotics and gaming, though it appears incremental as it builds on existing implicit policy methods with a novel architectural tweak.
The paper tackled the problem of improving reinforcement learning policy architectures by introducing Implicit Two-Tower (ITT) policies, which use attention scores between latent state and action representations, resulting in substantial computational gains and better performance compared to unstructured implicit and explicit policies across 15 environments.
We present a new class of structured reinforcement learning policy-architectures, Implicit Two-Tower (ITT) policies, where the actions are chosen based on the attention scores of their learnable latent representations with those of the input states. By explicitly disentangling action from state processing in the policy stack, we achieve two main goals: substantial computational gains and better performance. Our architectures are compatible with both: discrete and continuous action spaces. By conducting tests on 15 environments from OpenAI Gym and DeepMind Control Suite, we show that ITT-architectures are particularly suited for blackbox/evolutionary optimization and the corresponding policy training algorithms outperform their vanilla unstructured implicit counterparts as well as commonly used explicit policies. We complement our analysis by showing how techniques such as hashing and lazy tower updates, critically relying on the two-tower structure of ITTs, can be applied to obtain additional computational improvements.