Search papers, labs, and topics across Lattice.
Dream-MPC is introduced, a novel approach to model-based RL that combines a learned policy with gradient-based Model Predictive Control (MPC) using a learned world model. It addresses the challenge of gradient-based MPC often underperforming gradient-free methods by incorporating uncertainty regularization and amortizing optimization iterations. Experiments across 24 continuous control tasks demonstrate that Dream-MPC significantly improves policy performance, outperforming gradient-free MPC and state-of-the-art baselines.
Gradient-based MPC can finally beat gradient-free methods in continuous control, thanks to Dream-MPC's clever combination of learned policies, world models, uncertainty regularization, and optimization amortization.
State-of-the-art model-based Reinforcement Learning (RL) approaches either use gradient-free, population-based methods for planning, learned policy networks, or a combination of policy networks and planning. Hybrid approaches that combine Model Predictive Control (MPC) with a learned model and a policy prior to leverage the advantages of both paradigms have shown promising results. However, these approaches typically rely on gradient-free optimization methods, which can be computationally expensive for high-dimensional control tasks. While gradient-based methods are a promising alternative, recent works have empirically shown that gradient-based methods often perform worse than their gradient-free counterparts. We propose Dream-MPC, a novel approach that generates few candidate trajectories from a rolled-out policy and optimizes each trajectory by gradient ascent using a learned world model, uncertainty regularization and amortization of optimization iterations over time by reusing previously optimized actions. Our results on 24 continuous control tasks show that Dream-MPC can significantly improve the performance of the underlying policy and can outperform gradient-free MPC and state-of-the-art baselines. We will open source our code and more at https://dream-mpc.github.io.