Latent Diffusion Planning for Imitation Learning
This addresses the data efficiency problem in imitation learning for robotics, though it appears incremental as it builds on existing diffusion and latent space methods.
The paper tackles the problem of imitation learning's reliance on large amounts of expert demonstrations by proposing Latent Diffusion Planning (LDP), a modular approach that leverages action-free and suboptimal data through separate planning and inverse dynamics models in a learned latent space, resulting in outperforming state-of-the-art methods on simulated visual robotic manipulation tasks.
Recent progress in imitation learning has been enabled by policy architectures that scale to complex visuomotor tasks, multimodal distributions, and large datasets. However, these methods often rely on learning from large amount of expert demonstrations. To address these shortcomings, we propose Latent Diffusion Planning (LDP), a modular approach consisting of a planner which can leverage action-free demonstrations, and an inverse dynamics model which can leverage suboptimal data, that both operate over a learned latent space. First, we learn a compact latent space through a variational autoencoder, enabling effective forecasting of future states in image-based domains. Then, we train a planner and an inverse dynamics model with diffusion objectives. By separating planning from action prediction, LDP can benefit from the denser supervision signals of suboptimal and action-free data. On simulated visual robotic manipulation tasks, LDP outperforms state-of-the-art imitation learning approaches, as they cannot leverage such additional data.