S2P: State-conditioned Image Synthesis for Data Augmentation in Offline Reinforcement Learning
This addresses a bottleneck in offline RL for image-based tasks, offering a novel method to bridge state and image domains, though it is incremental as it builds on existing state-based approaches.
The paper tackles the problem of distributional shift in image-based offline reinforcement learning by proposing S2P, a generative model that synthesizes raw pixel images from states, enabling virtual exploration and data augmentation. The result shows improved performance and generalization on unseen tasks in experiments.
Offline reinforcement learning (Offline RL) suffers from the innate distributional shift as it cannot interact with the physical environment during training. To alleviate such limitation, state-based offline RL leverages a learned dynamics model from the logged experience and augments the predicted state transition to extend the data distribution. For exploiting such benefit also on the image-based RL, we firstly propose a generative model, S2P (State2Pixel), which synthesizes the raw pixel of the agent from its corresponding state. It enables bridging the gap between the state and the image domain in RL algorithms, and virtually exploring unseen image distribution via model-based transition in the state space. Through experiments, we confirm that our S2P-based image synthesis not only improves the image-based offline RL performance but also shows powerful generalization capability on unseen tasks.