Timezone: »

 
Model-Based Offline Planning with Trajectory Pruning
Xianyuan Zhan · Xiangyu Zhu · Haoran Xu

@
in Offline Reinforcement Learning »

Offline reinforcement learning (RL) enables learning policies using pre-collected datasets without environment interaction, which provides a promising direction to make RL usable in real-world systems. Although recent offline RL studies have achieved much progress, existing methods still face many practical challenges in real-world system control tasks, such as computational restriction during agent training and the requirement of extra control flexibility. Model-based planning framework provides an attractive solution for such tasks. However, most model-based planning algorithms are not designed for offline settings. Simply combining the ingredients of offline RL with existing methods either provides over-restrictive planning or leads to inferior performance. We propose a new light-weighted model-based offline planning framework, namely MOPP, which tackles the dilemma between the restrictions of offline learning and high-performance planning. MOPP encourages more aggressive trajectory rollout guided by the behavior policy learned from data, and prunes out problematic trajectories to avoid potential out-of-distribution samples. Experimental results show that MOPP provides competitive performance compared with existing model-based offline planning and RL approaches.

Author Information

Xianyuan Zhan (Tsinghua University, Tsinghua University)
Xiangyu Zhu (Tianjin University)
Haoran Xu (JD Technology)

More from the Same Authors