Offline Reinforcement Learning as Anti-Exploration
This addresses the challenge of learning optimal control from fixed datasets without system interactions, which is crucial for applications where online exploration is costly or unsafe, though it is an incremental improvement over existing methods.
The paper tackled the problem of offline reinforcement learning by proposing an agent that subtracts a prediction-based exploration bonus from rewards to keep the policy close to the dataset support, showing it is competitive with state-of-the-art methods on continuous control tasks.
Offline Reinforcement Learning (RL) aims at learning an optimal control from a fixed dataset, without interactions with the system. An agent in this setting should avoid selecting actions whose consequences cannot be predicted from the data. This is the converse of exploration in RL, which favors such actions. We thus take inspiration from the literature on bonus-based exploration to design a new offline RL agent. The core idea is to subtract a prediction-based exploration bonus from the reward, instead of adding it for exploration. This allows the policy to stay close to the support of the dataset. We connect this approach to a more common regularization of the learned policy towards the data. Instantiated with a bonus based on the prediction error of a variational autoencoder, we show that our agent is competitive with the state of the art on a set of continuous control locomotion and manipulation tasks.