We propose a novel approach for automatically generating a move plan forscene arrangement.1 Given a scene like an apartment with many furnitureobjects, to transform its layout into another layout, one would need to determinea collision-free move plan. It could be challenging to design this planmanually because the furniture objects may block the way of each other ifnot moved properly; and there is a large complex search space of move actionsequences that grow exponentially with the number of objects. To tacklethis challenge, we propose a learning-based approach to generate a moveplan automatically. At the core of our approach is a Monte Carlo tree thatencodes possible states of the layout, based on which a search is performed to move a furniture object appropriately in the current layout. We trained apolicy neural network embedded with a LSTM module for estimating thebest actions to take in the expansion step and simulation step of the MonteCarlo tree search process. Leveraging the power of deep reinforcement learning,the network learned how to make such estimations through millions oftrials of moving objects. We demonstrated our approach for moving objectsunder different scenarios and constraints. We also evaluated our approachon synthetic and real-world layouts, comparing its performance with thatof humans and other baseline approaches.
展开▼