Causal Models as a Framework for Human-Robot Interaction
Abstract
Approved for Public Release -- In the future, physical systems from simple tools to complex machines will be built, used, and repair,ed by humans and robots working together. Joint work requires a robot to perceive, understand, and share the semantics of its enviro,nment with a human partner. A robot must be able to interpret a users commands in a context that accords with that users goals and, model of the world. In turn, the robot must be able to communicate facts about the world and its intentions in ways that are both p,ertinent and comprehensible to the human. This project will explore how interaction with humans can be leveraged to facilitate robo,t learning of causal models to support action. Causal models in the form of modified Causal Bayes Nets will be used to represent ca,usal systems and provide a shared representation for human and robot. The idea is to determine how invariant relationships captured, from causal relations can be used to aid in executing robot tasks. A central issue we wish to evaluate is how well the causal info,rmation serves to support generalized learning that can carry over to different problems. We will accomplish this evaluation by set,ting up a series of experiments where a robot has a collection of causal models of actions and of the structure of objects and is th,en asked to do an assembly task with components of varying size and structure or a troubleshooting task with non-functioning objects,. We hypothesize that this causal structure will provide robustness and a generalization capability beyond that of associational str,uctures. The findings of this project will assist in the design of autonomous robots that can use human-provided causal information, to complete a wide-range of tasks not encountered before, including in inhospitable environments. The technologies developed throu,gh this proposal will be of use to both the military and civilian applications.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Jul 13, 2022
- Source ID
- N000142212494
Entities
People
- Ruth Bahar
Organizations
- Colorado School of Mines
- Office of Naval Research
- United States Navy