Heuristic search has traditionally relied on hand-crafted or programmatically
derived heuristics. Neural networks (NNs) are newer powerful tools which can be
used to learn complex mappings from states to cost-to-go heuristics. However,
their slow single inference time is a large overhead that can substantially
slow down planning time in optimized heuristic search implementations. Several
recent works have described ways to take advantage of NN's batch computations
to decrease overhead in planning, while retaining bounds on (sub)optimality.
However, all these methods have used the NN heuristic in a "blocking" manner
while building up their batches, and have ignored possible fast-to-compute
admissible heuristics (e.g. existing classically derived heuristics) that are
usually available to use. We introduce Non-Blocking Batch A* (NBBA*), a bounded
suboptimal method which lazily computes the NN heuristic in batches while
allowing expansions informed by a non-NN heuristic. We show how this subtle but
important change can lead to substantial reductions in expansions compared to
the current blocking alternative, and see that the performance is related to
the information difference between the batch computed NN and fast non-NN
heuristic.

使用神经网络进行启发式搜索的研究中，Non-Blocking Batch A* 提出了一种新的边界次优方法，该方法可以懒惰地以批处理方式计算神经网络启发式搜索结果，同时使用非神经网络启发式搜索算法进行扩展。在实验中发现，相较于当前阻塞式替代方法，这种微妙但重要的变化可以显著减少扩展次数，而性能与批处理计算的神经网络和快速的非神经网络启发式算法之间的信息差有关。