Combining Learning and Reasoning - Towards Human-Level Robot Intelligence

Organizers: Peter Karkus, Alina Kloss, Rico Jonschkowski, Leslie P. Kaelbling


Robotics research has developed powerful model-based methods for perception, state estimation, planning, control, etc., which form the building blocks of the vast majority of successful robot systems. At the same time, data-driven, model-free learning has recently brought unprecedented success in various domains, where model-based methods struggle despite decades of research. The aim of this workshop is to bring together researchers from robotics and machine learning, and discuss opportunities and challenges towards building human-level robot intelligence, in particular, combining model-based reasoning and data-driven learning in a scalable and composable manner.

Some questions we would like to discuss:

  • What is the role of model-based reasoning and model-free learning in an intelligent robot system?
  • Can we learn intelligent robot behaviour only from reinforcements? Is expert knowledge essential?
  • How do we combine existing knowledge with data-driven learning? What should be built in and what should be learned?
  • How do we integrate solutions for small, isolated sub-tasks into a large intelligent system? How do we combine model-based and model-free components? Would we build Shakey differently today?
  • Is robot intelligence going to be explainable? Is interpretability unnecessary, good to have, or a must?