BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:Message-Aware Graph Attention Networks for Large-Scale Multi-Robot
  Path Planning - Qingbiao Li
DTSTART:20210316T131500Z
DTEND:20210316T141500Z
UID:TALK156760@talks.cam.ac.uk
CONTACT:Mateja Jamnik
DESCRIPTION:"Join us on Zoom":https://zoom.us/j/99166955895?pwd=SzI0M3pMVE
 kvNmw3Q0dqNDVRalZvdz09\n\nThe domains of transport and logistics are incre
 asingly relying on autonomous mobile robots for the handling and distribut
 ion of passengers or resources. At large system scales\, finding decentral
 ized path planning and coordination solutions is key to efficient system p
 erformance. Recently\, Graph Neural Networks (GNNs) have become popular du
 e to their ability to learn communication policies in decentralized multi-
 agent systems. Yet\, vanilla GNNs rely on simplistic message aggregation m
 echanisms that prevent agents from prioritizing important information.\nTo
  tackle this challenge\, in this paper\, we extend our previous work that 
 utilizes GNNs in multi-agent path planning by incorporating a novel mechan
 ism to allow for message-dependent attention. Our Message-Aware Graph Atte
 ntion neTwork (MAGAT) is based on a key-query-like mechanism that determin
 es the relative importance of features in the messages received from vario
 us neighboring robots. We show that MAGAT is able to achieve a performance
  close to that of a coupled centralized expert algorithm. Further\, ablati
 on studies and comparisons to several benchmark models show that our atten
 tion mechanism is very effective across different robot densities and perf
 orms stably in different constraints in communication bandwidth.\nExperime
 nts demonstrate that our model is able to generalize well in previously un
 seen problem instances\, and it achieves a 47% improvement over the benchm
 ark success rate\, even in very large-scale instances that are 100x larger
  than the training instances.
LOCATION:Zoom
END:VEVENT
END:VCALENDAR