ICRA2017 recently took place in Singapore. As usual, it was accompanied by a constellation of workshops and special meetings including an organizing meeting our our ICRA2019 group, a very nice workshop on self-driving cars, and a satellite meeting at NUS on marine robotics. ICRA is the flagship conference of the robotics and automation society and the premiere venue for robotics researchers to exchange ideas, results, news, technologies and generally do networking. The meeting had over 2,400 attendees which reflects the growing shift from pure long-term academic research to immediate socio-economic impact. Even though this is a research meeting, not an industrial development meeting (like RoboBusiness), it includes a growing trade show, a lot of industrial engagement, and an increasingly hot recruiting component.
While the main research themes apparent at the conference were largely predictable outgrowths of ongoing trends, there are continuing shifts in emphasis that I noticed.
Notably, soft robotics is growing in impact and has some interesting implications for manipulation. Self-driving vehicles are such an obvious theme that they are hardly worth mentioning as a "new theme", except to observe there is a bit of a bifurcation between some of the long term research in the domain and some of the more pragmatic stuff that is not being exposed in public due to its commercial impact and potential. At the self-driving car workshop, for example, there was a notable lack of technical details in all the presentations from commercially-driven researchers. While deep learning is making significant inroads and having an impact on robotics, its effects are more nuanced than in, say computer vision, due to the importance of physical embodiment, system integration and the diverse set of issues that distinguish robotics from man other research disciplines. The majority of good university labs, including ours, are working at the intersection of robotics and deep learning, but there was a lot of exciting research that didn't need to invoke that technology at all, or only as a small black box. One of my favorite sub-themes was related to policy transfer and policy sharing. Another theme I thought was interesting and rapidly maturing, but which we don't work on, was event cameras.
Notably, two important figures and friends of mine won major awards for their ongoing work: Oussama Khatib from Stanford and Vijay Kumar from Penn. Both of these researchers have made, and continue to make, major impact on the field it is was great to see them recognized. Best paper awards went to Kostas Daniilidis and his coauthors from U Penn on Semantic SLAM and to Dieter Fox (U Washington) and his co-authors (also from Oculus) for "Self-supervised Visual Descriptor Learning for Dense Correspondence". My student Jimmy Li also works on aspects semantic SLAM and its both exciting and a bit stressful to see the rapid progress in the field.
Of course, with hundreds of results being presented as well as a diverse set of side meetings, my representation of the conference is obviously very subjective and unavoidably incomplete. Overall the paper quality seems good, the organization was superb and the venues was beautiful. Singapore was an impressive place with fabulous infrastructure and an exciting mix of cultures.
Here I am in Singapore
Award ceremony for
Oussama Khatib (Stanford)