Assembling furniture amounts to solving the discrete-continuous optimization task of selecting the furniture parts to assemble and estimating their connecting poses in a physically realistic manner. The problem is hampered by its combinatorially large yet sparse solution space thus making learning to assemble a challenging task for current machine learning models. In this paper, we attempt to solve this task by leveraging the assembly instructions provided in diagrammatic manuals that typically accompany the furniture parts. Our key insight is to use the cues in these diagrams to split the problem into discrete and continuous phases. Specifically, we present Manual-PA, a transformer-based instruction Manual-guided 3D Part Assembly framework that learns to semantically align 3D parts with their illustrations in the manuals using a contrastive learning backbone towards predicting the assembly order and infers the 6D pose of each part via relating it to the final furniture depicted in the manual. To validate the efficacy of our method, we conduct experiments on the benchmark PartNet dataset. Our results show that using the diagrams and the order of the parts lead to significant improvements in assembly performance against the state of the art. Further, Manual-PA demonstrates strong generalization to real-world IKEA furniture assembly on the IKEA-Manual dataset.

A fixed schedule dynamic network has a set of nodes (eg. vehicles or satellites) that move using a known schedule and trajectory, so that the connections between nodes in the network appear and disappear in a predictable manner. We study the problem of finding a foremost journey in such a network: given a query time, source and destination node, we find a temporal path that arrives at the earliest time at the destination. We give a new approach to the problem that uses a timetable sorted in order of the start time of connections, and describe three algorithms using this approach. We prove their correctness and give tight bounds on their worst-case time complexities. We also show extensive experimental results that show that our new algorithms outperform the previous best algorithm given in [1].