Double Horizon Model-Based Policy Optimization

TL;DR

The paper introduces DHMBPO, a novel model-based reinforcement learning method that employs two different rollout horizons to better balance bias, variance, and distribution shift, leading to improved efficiency and stability.

Contribution

It proposes a double-horizon approach dividing rollouts into distribution and training phases, addressing conflicting optimal horizons in model-based RL.

Findings

Outperforms existing MBRL methods on continuous-control benchmarks.

Achieves higher sample efficiency and lower runtime.

Effectively balances distribution shift, model bias, and gradient variance.

Abstract

Model-based reinforcement learning (MBRL) reduces the cost of real-environment sampling by generating synthetic trajectories (called rollouts) from a learned dynamics model. However, choosing the length of the rollouts poses two dilemmas: (1) Longer rollouts better preserve on-policy training but amplify model bias, indicating the need for an intermediate horizon to mitigate distribution shift (i.e., the gap between on-policy and past off-policy samples). (2) Moreover, a longer model rollout may reduce value estimation bias but raise the variance of policy gradients due to backpropagation through multiple steps, implying another intermediate horizon for stable gradient estimates. However, these two optimal horizons may differ. To resolve this conflict, we propose Double Horizon Model-Based Policy Optimization (DHMBPO), which divides the rollout procedure into a long "distribution rollout" (DR) and a short "training rollout" (TR). The DR generates on-policy state samples for mitigating distribution shift. In contrast, the short TR leverages differentiable transitions to offer accurate value gradient estimation with stable gradient updates, thereby requiring fewer updates and reducing overall runtime. We demonstrate that the double-horizon approach effectively balances distribution shift, model bias, and gradient instability, and surpasses existing MBRL methods on continuous-control benchmarks in terms of both sample efficiency and runtime.