BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:Graph neural network approach for decentralized multi-robot coordi
 nation - University of Cambridge
DTSTART:20220421T100000Z
DTEND:20220421T110000Z
UID:TALK173057@talks.cam.ac.uk
CONTACT:Dr H Ge
DESCRIPTION:Effective communication is key to successful\, decentralized\,
 \nmulti-robot path planning. Yet\, it is far from obvious what\ninformatio
 n is crucial to the task at hand\, and how and when it must\nbe shared amo
 ng robots. To side-step these issues and move beyond\nhand-crafted heurist
 ics\, we propose a combined model that\nautomatically synthesizes local co
 mmunication and decision-making\npolicies for robots navigating in constra
 ined workspaces. Our\narchitecture is composed of a convolutional neural n
 etwork (CNN) that\nextracts adequate features from local observations\, an
 d a graph neural\nnetwork (GNN) that communicates these features among rob
 ots. We train\nthe model to imitate an expert algorithm\, and use the resu
 lting model\nonline in decentralized planning involving only local communi
 cation\nand local observations. We evaluate our method in simulations by\n
 navigating teams of robots to their destinations in 2D cluttered\nworkspac
 es. We measure the success rates and sum of costs over the\nplanned paths.
  The results show a performance close to that of our\nexpert algorithm\, d
 emonstrating the validity of our approach. In\nparticular\, we show our mo
 del's capability to generalize to previously\nunseen cases (involving larg
 er environments and larger robot teams).\nIn Today's talk\, the speaker wi
 ll present his work from proof of\nconcept in simulation into the real-wor
 ld robotics systems toward the\nfully decentralized system.\n\n\n\nSpeaker
 's Bio:\n\nQingbiao Li is a final year PhD student at Prorok Lab in the Di
 gital\nTechnology Group (DTG) at the University of Cambridge under the\nsu
 pervision of Dr Amanda Prorok. During his PhD\, he focuses on\ncommunicati
 on-aware motion planning for multi-robot coordination. He\nis investigatin
 g Graph Neural Networks (GNN) to build communication\nchannels for multi-a
 gent and multi-robot systems so that they can\nlearn how to communicate be
 tween each other explicitly. His research\ncan be applied to mobility-on-d
 emand systems\, automated warehouses\,\nand smart cities.\n\nPrior to join
 ing Cambridge\, he was working in the Hamlyn Centre at\nImperial College L
 ondon\, founded by Prof Guang-Zhong Yang and the Lord\nAra Darzi\, in the 
 field of medical robotics and healthcare to earn my\nMRes in Medical Robot
 ics and Image-Guided Intervention. He was\nsupervised by Prof Daniel Elson
  for an eight-month research project\nabout oxygen saturation (StO2) estim
 ation\, and graduated with a\ndistinction. He also held an MEng degree in 
 Mechanical Engineering\nfrom the University of Edinburgh.\n\nHomepage:  ht
 tp://qingbiaoli.github.io/
LOCATION:Hybrid meeting\, CBL seminar room\, and Zoom https://eng-cam.zoom
 .us/j/87988930234?pwd=eHRGTHFtUE9UL0pJOWNtdDF2VDN2UT09
END:VEVENT
END:VCALENDAR