Abstract

Even though many problems of interest involve trade-offs between opposing criteria, direct treatment as multi-objective problems remains relatively rare in our field. This workshop is being proposed on the basis of three possible ideas for why that might be: (1) general obliviousness; (2) a sense of inapplicability; and (3) inaccessibility.  We are seeking to bring researchers together to explore which of these ideas hold up under closer scrutiny, examine their implications, and propose possible remedies.  The workshop seeks to draw not only on people who have a track record with multi-objective optimization, but also researchers in adjacent areas or those in suitable robotics applications. 

Topics of interest include, but are not limited to:

• Multi-objective optimization and planning in robotics

• Balancing trade-offs in planning

• Parameter tuning / auto-tuning of controllers and their objectives

• Active perception

• Reward design / designing cost functions

• Multi-objective RL

• Preference learning and personalization of robots

• Inverse reinforcement learning / Inverse optimal control

• Robot design