BEGIN:VCALENDAR
VERSION:2.0
PRODID:Linklings LLC
BEGIN:VTIMEZONE
TZID:Asia/Singapore
X-LIC-LOCATION:Asia/Singapore
BEGIN:STANDARD
TZOFFSETFROM:+0800
TZOFFSETTO:+0800
TZNAME:SGT
DTSTART:19820101T123000
END:STANDARD
END:VTIMEZONE
BEGIN:VEVENT
DTSTAMP:20201212T050115Z
LOCATION:Zoom Room 7
DTSTART;TZID=Asia/Singapore:20201212T110000
DTEND;TZID=Asia/Singapore:20201212T110500
UID:siggraphasia_SIGGRAPH Asia 2020_sess111_papers_198@linklings.com
SUMMARY:TAP-Net: Transport-and-Pack using Reinforcement Learning
DESCRIPTION:Technical Papers, Technical Papers Q&A\n\nTAP-Net: Transport-a
nd-Pack using Reinforcement Learning\n\nHu, Xu, Chen, Gong, Zhang...\n\nWe
introduce the transport-and-pack problem and develop a neural optimizatio
n model to solve it based on reinforcement learning. Given an initial spat
ial configuration of boxes, we seek an efficient method to iteratively tra
nsport and pack the boxes compactly into a target container. In general, p
acking alone is a well-known, difficult combinatorial optimization problem
. Due to obstruction and accessibility constraints, our problem has to add
a transport planning dimension to the already immense search space. Using
a learning-based approach, a trained network can learn and encode solutio
n patterns to guide the solution of new problem instances instead of execu
ting an expensive online search. In our work, we represent the various con
straints using a precedence graph and train a neural network, coined TAP-N
et, using reinforcement learning to reward efficient packing. The network
is built on a recurrent neural network (RNN) which inputs the current prec
edence graph, as well as the current box packing state of the target conta
iner, and it outputs the next box to pack, as well as its orientation. We
train our network on randomly generated initial box configurations, withou
t supervision, via policy gradients to learn optimal TAP policies to maxim
ize packing efficiency. We demonstrate the performance of TAP-Net on a var
iety of examples, evaluating the network through ablation studies and comp
arisons to baselines and heuristic search methods. We also show that our n
etwork generalizes well to larger problem instances, when trained on small
-sized inputs.\n\nRegistration Category: Ultimate Supporter, Ultimate Atte
ndee
END:VEVENT
END:VCALENDAR