Relational activity processes for modeling concurrent cooperation

Abstract

In human-robot collaboration, multi-agent domains, or single-robot manipulation with multiple end-effectors, the activities of the involved parties are naturally concurrent. Such domains are also naturally relational as they involve objects, multiple agents, and models should generalize over objects and agents. We propose a novel formalization of relational concurrent activity processes that allows us to transfer methods from standard relational MDPs, such as Monte-Carlo planning and learning from demonstration, to concurrent cooperation domains. We formally compare the formulation to previous propositional models of concurrent decision making and demonstrate planning and learning from demonstration methods on a real-world human-robot assembly task.

Impact of Robot Initiative on Human-Robot Collaboration

“…The pick and give actions are realized with the left arm while the hold action is realized with the right one. The robot is controlled using the RAP framework [7] that allows concurrently running actions and uses relational representations for states and actions.…”

Flexible human–robot cooperation models for assisted shop-floor tasks

“…Its application to human-robot cooperation tasks (F 3 ) has not been fully addressed in the literature. An approach in that direction is described in [25], where an integrated approach to Monte Carlo based action planning and trajectory planning via Programming by Demonstration is adopted in a scenario of toolbox assembly. Concurrent activities are formalized using a Markov decision process, which determines when initiating and terminating each human or robot action.…”

Preference learning on the execution of collaborative human-robot tasks

“…In this section we present the underlying formalism to model team behavior. We rely on a recently introduced general formalism for relational MDPs with concurrent actions [1], and show how to apply it to team behavior.…”

“…A Relational Activity Processes (RAP), introduced in [1], is a way to model concurrent cooperation of multiple agents while being at the same time a Markovian process. Roughly, RAPs define a sequential MDP where decisions are about the initiation or termination of activities, and activities run concurrently with random durations.…”

“…In this paper we propose different ways to learn the preferences over the execution of a collaborative task with concurrent actions. Our contributions include the following: i) a formalism to model team behavior using Relational Activity Processes (RAPs) [1]; ii) a formalism to represent preferences with standard MDP tools and iii) two different algorithms to learn preferences interactively. We show results in simulation and with a real robot.…”