Workshop on the Integration of Control Theory, Formal Methods, Learning and Human Factors for Autonomous Systems

From u-t-autonomous.info
Jump to: navigation, search

Date: April 18-19, 2016
Location: The University of Texas at Austin, The Peter O'Donnell, Jr. Building, POB 2.402 (location of the building on Google maps)
Objective: Each of the disciplines, control theory, formal methods, learning and human factors, has the potential to contribute toward affordable, trustworthy and user-friendly autonomous systems. On the other hand, the existing disciplinary boundaries often limit their impact. For example, while control theory and formal methods offer rigorous proofs of correctness with respect to rich sets of mathematically based specifications, they lack capabilities for adaptation to contextual changes. Learning offers such adaptability yet often at the expense of clear semantics for robustness and flexible high-level specifications. The disconnect of control theory, formal methods and learning from the growing human factors and modeling research complicates their acceptance by humans. We believe that tighter interactions between these disciplines is necessary. The workshop will emphasize the development of hybrid methods that blend ideas from all these disciplines and possibly others toward a unified language and toolset.
Organizers: Michael Goodrich (BYU), Laura Humphrey (AFRL), Michael Littman (Brown), and Ufuk Topcu (UT Austin)
Acknowledgements: The workshop is partly funded by AFRL and ICES. Thanks to Cynthia Steiner, Murat Cubuktepe, Deep Mukherji and Suda Bharadwaj for local arrangements.

Participants

  • Julie Adams (Vanderbilt University)
  • Amy Pritchett (Georgia Institute of Technology)
  • Matthew Bolton (University of Buffalo)
  • Chris Miller (Smart Information Flow Technologies)
  • Nisar Ahmed (University of Colorado)
  • John Laird (University of Michigan)
  • Anca Dragan (University of California, Berkeley)
  • Marta Kwiatkowska (University of Oxford)
  • Moshe Vardi (Rice University)
  • Jose del Millan (Swiss Federal Institute of Technology in Lausanne)
  • Jie Fu (Worcester Polytechnic Institute)
  • Andrea Thomaz (The University of Texas at Austin)
  • Ruzena Bajcys (University of California, Berkeley)
  • Matthias Scheutz (Tufts University)
  • Bill Smart (Oregon State University)
  • Shie Mannor (Technion)
  • Greg Trafton (Naval Research Laboratory)
  • Peter Trautman (Galois)
  • Kris Kearns (AFRL)
  • Irene Gregory (NASA)
  • Derek Kingston (AFRL)
  • Peter Stone (The University of Texas at Austin)
  • Angela Schoellig (University of Toronto)
  • Michael Goodrich (Brigham Young University)
  • Michael Littman (Brown University)
  • Laura Humphrey (AFRL)
  • Ufuk Topcu (The University of Texas at Austin)
  • Nils Jansen (The University of Texas at Austin)
  • Julia Badger (NASA)
  • Scott Niekum (The University of Texas at Austin)
  • Min Wen (University of Pennsylvania)
  • Salar Moarref (University of Pennsylvania)

Schedule

April 18th, 2016

  • 8:30-8:40 -- Introduction
  • 8:40-11:50 -- Overview talks [30 minutes for each talk followed by 10 minutes of discussion]
    • 8:40-9:20 -- Marta Kwiatkowska and Moshe Vardi (formal methods) -- Vardi's slides
    • 9:20-10:00 -- Andrea Thomaz and Peter Stone (learning)
    • 10:30-11:10 -- Ruzena Bajcsy and John Laird (control)
    • 11:10-11:50 -- Amy Pritchett and Nisar Ahmed (human factors)

Each talk will provide an overview of what the respective discipline has to offer toward developing better autonomous systems. The talks will be structured about the following (tentative list of questions).

  1. How is autonomy interpreted in your field?
  2. What are key problems your field thinks about?
  3. What are some problems that you view as solved?
  4. What are some problems that you view as not solved but important?
  5. What is a technical idea that might be powerful and general enough that it might be of use by others?

The 10-minute discussion period will be structured around the reactions of the audience on the questions above and the comments made by the speaker.

  • 1:15-2:15 Lightning talks --- One-slide-ninety-second talks by the participants regarding the following questions.
    • What are your name, affiliation and main research area?
    • What is *one* of your most exciting (to you) research results that is relevant for this workshop? (Or, one lesson you learned, one project you have, one failure you had, etc. about which you want to tell to this audience.)
    • What is *the* question you want to be discussed during the break-out sessions or you hope to be answered by one of the participants (feel free to name people)?
  • 2:45-4:45 -- Challenge the other side or “reveal your ignorance” [15 minutes for each presentation followed by 10 minutes for reactions] For each of the other subareas at the meeting, name a problem that you believe that subarea might be able to help you with.
    • 2:45-3:15 -- Matthew Bolton and Ufuk Topcu (challenge from a formal methods perspective)
    • 3:15-3:45 -- Shie Mannor and Michael Littman (challenge from a learning perspective)
    • 3:45-4:15 -- Irene Gregory and Derek Kingston (challenge from a controls perspective)
    • 4:15-4:45 -- Julie Adams and Matthias Scheutz (challenge from a human factors perspective)
  • 4:45-5 -- Wrap-up and plans and assignments for the second day

April 19th, 2016

  • 8:30-9:00 -- Reflections on the first day (what was cool? what did you not understand? what did you disagree with?)

A combination of brief talks by a subset of participants and open discussion.

  • 9:00-10:30 -- Application perspective on autonomy (what do we want autonomy to do? what is missing?)
    • 9:00-9:15 -- Kris Kearns
    • 9:15-9:30 -- Laura Humphrey
    • 9:30-9:45 -- Greg Trafton
    • 9:45-10:00 -- Chris Miller
    • 10:00-10:15 -- Bill Smart
    • 10:15-10:30 -- Jose del Millan
  • 11-11:15 -- Introduction of the research challenges for the working sessions
    • Anca Dragan and Nils Jansen
    • Angela Schoellig (?) and Laura Humphrey
    • Peter Trautman and Michael Goodrich

During the rest of the workshop, multiple sub-groups will discuss pre-specified research problems (one problem for each sub-group). The outcome of the sessions will be a draft of the abstract of a potential paper that may be later written on the problem.

  • 11:15-12 -- First part of the working sessions
  • 1:30-2:30 -- Second of the working sessions
  • 3-4 -- Discussion of the "paper" abstracts
  • 4-4:15 -- Wrap-up

Accommodation

  • The most convenient option is the AT&T Center on campus. It can get sold out quickly.
  • The map at this URL shows a number of hotels around the campus. Most are in the band between downtown and the campus. They are accessible by bus, taxis and within a reasonable walking distance.
  • Many of the recent visitors had quite a bit of luck on airbnb.com for short-term accommodation nearby the campus. Another popular option is this B&B next to the campus.
  • Austin Airport is [uber.com Uber] (and perhaps other ridesharing options) friendly.