Latent Plan Transformer for Trajectory Abstraction: Planning as Latent Space Inference
This addresses planning problems in offline reinforcement learning by providing a method for improved decision-making from sub-optimal data, though it is incremental as it builds on existing generative modeling approaches.
The paper tackles the challenge of planning in long-term tasks by introducing the Latent Plan Transformer (LPT), which uses latent variable inference to connect trajectory generation with final returns, achieving competitive performance on benchmarks like Gym-Mujoco and Maze2D.
In tasks aiming for long-term returns, planning becomes essential. We study generative modeling for planning with datasets repurposed from offline reinforcement learning. Specifically, we identify temporal consistency in the absence of step-wise rewards as one key technical challenge. We introduce the Latent Plan Transformer (LPT), a novel model that leverages a latent variable to connect a Transformer-based trajectory generator and the final return. LPT can be learned with maximum likelihood estimation on trajectory-return pairs. In learning, posterior sampling of the latent variable naturally integrates sub-trajectories to form a consistent abstraction despite the finite context. At test time, the latent variable is inferred from an expected return before policy execution, realizing the idea of planning as inference. Our experiments demonstrate that LPT can discover improved decisions from sub-optimal trajectories, achieving competitive performance across several benchmarks, including Gym-Mujoco, Franka Kitchen, Maze2D, and Connect Four. It exhibits capabilities in nuanced credit assignments, trajectory stitching, and adaptation to environmental contingencies. These results validate that latent variable inference can be a strong alternative to step-wise reward prompting.