The What without the How: Specifying Planning Problems in Robotics
Every new graduate student is told that when it comes to research, a good problem formulation is half the battle won. As the responsibilities we give to robots and autonomous vehicles become more complex by the day, the time is ripe to step back and redefine the basic blocks that define the planning problems we solve when creating “autonomy”. The classical piano-mover’s problem has dominated the motion planning literature for decades, and is still unsolved for many domains, but increasingly, planning problems involve responding to uncertainty, stochasticity, partial observability, unwieldy dynamics, and other challenges. Moreover, the “what” itself is evolving. Planning for long-term autonomy, collaborative and surveillance tasks have all motivated unique reformulations of the classical planning problem, including novel means of specifying desired behavior such as temporal logics and domain-specific languages, as well as natural language instructions. This workshop is an attempt to wrangle these disparate approaches and make some sense of the zoo of options. Participants will be treated to a breadth of current approaches from area experts, and encouraged to be critical of new modes of specification and the utility of the planning approaches that go therewith.
Description of the event
The first half of the workshop will feature a series of warm-up talks to expose attendees to the wide variety of problem formulations that are currently used to solve planning problems in robotics. Each speaker will have ~20 minutes (including questions) to describe a problem and its solution, with a focus on how the specification was crucial to the approach. We have proposed speakers who can speak to various aspects and flavors of the planning problem.
For the second half of the workshop, we will solicit contributed lightning talks of ~5 min of the following format: propose two formulations of the same planning problem (i.e. with different specification languages/modes) and argue for one.
In the final 1.5 hours of the workshop, we will play a party-planning game. We will divide the participants into an even number of teams. Each team will be randomly assigned a (party-themed) task, and charged with coming up with a specification for the task. The idea is to specify the task so well that another team can plan for and execute it correctly. Teams will then swap task specifications with a different team, and come up with a plan for achieving the task as specified, but do so in troll-mode, i.e. adversarially. In other words, upon receiving the other team’s problem description, each team’s goal is to solve the problem, but in the “worst” way possible. The goal of this activity is to demonstrate the non-triviality of well-specified planning problems. We will wrap up with an open discussion of lessons learned, via the game, as well as throughout the day.
We will then have a (hopefully not horrible) party.
Contact the organizers: firstname.lastname@example.org, email@example.com
- Workshop Date: Saturday July 15th 2017, Location: 32-155
- Abstract Submission Deadline: May 19th 2017
- Acceptance Notifications: June 11th 2017
- Presentation Submission Deadline: July 10th 2017
Call For Lightning Talks
We solicit abstracts (~ 1 page) on specification of planning problems, the relationship between specification and planning, design choices, and challenges. To foster critical discussions, abstracts should consider a problem that can be formulated and solved in two different ways. Moreover, you should chose one formulation and argue for it. You should focus on what was specified, what was planned for, and the advantages and disadvantages. We ask that the abstract have the following format: 1) informal problem description; 2) formulation 1 and solution; 3) formulation 2 and solution; 4) argument for either design choice.
Abstracts should be send by email to firstname.lastname@example.org and are due by May 15th 2017. The emails must have the subject “[RSS2017-SPR] Abstract submission”.
Selected abstracts will be presented at the workshop PechaKucha-style: 15 slides over 5 minutes (20 seconds per slide). Final presentations must be submitted by July 10th 2017 in Powerpoint with standard (4:3) format.
- Prof. Calin Belta, Boston University, USA
- Dr. Andrea Censi, ETH Zürich, Switzerland
- Prof. Leslie Kaelbling, Massachusetts Institute of Technology, USA
- Prof. George Konidaris, Brown University, USA
- Prof. Hadas Kress-Gazit, Cornell University, USA
- Prof. Amy LaViers, University of Illinois at Urbana-Champaign, USA
- Prof. Daniele Magazzeni, King’s College London, UK
- Prof. Jason O’Kane, University of South Carolina, USA
- Prof. Stefanie Tellex, Brown University, USA
|8:45 – 9:00||Welcome and Workshop Overview|
|Session 1: 9:00 – 10:30||4 talks|
|09:00 – 09:20||Speaker: Calin Belta|
|09:20 – 9:40||Speaker: Leslie Kaelbling|
|9:40 – 10:00||Speaker: George Konidaris|
|10:00 – 10:20||Speaker: Jason O’Kane|
|10:30 – 11:00||Coffee Break|
|Session 2: 11:00 – 12:00||3 talks|
|11:00 – 11:20||Speaker: Hadas Kress-Gazit|
|11:20 – 11:40||Speaker: Cristian-Ioan Vasile|
|11:40 – 12:00||Speaker: Daniele Magazzeni|
|12:00 – 14:00||Lunch break|
|Session 3: 14:00 – 14:45||2 talks|
|14:00 – 14:20||Speaker: Amy LaViers|
|14:20 – 14:40||Speaker: Stefanie Tellex|
|Moderated Session: 14:40 – 15:00||Conclusions, Problems, Research Directions|
|15:00 – 15:30||Coffee Break|
|Lightning Session: 15:30 – 16:00||5 talks|
|15:30 – 15:35||Speaker: Dale McConachie|
|15:35 – 15:40||Speaker: Schillinger Philipp|
|15:40 – 15:45||Speaker: Maira Saboia Da Silva|
|15:45 – 15:50||Speaker: Cenk Baykal|
|15:50 – 15:55||Speaker: Brandon Araki|
|Interactive game: 16:00 – 17:30|