Goal-Directed Reasoning and Cooperation in Robots in Shared Workspaces: An Internal Simulation-Based Neural Framework

Abstract

In an industrial assembly task, two robots share workspace to assemble fuse-boxes. We describe coupled forward models of body and peripersonal space learned via exploration. Internal simulation evaluates action-consequence pairs, driving cooperative sequences even when solo goals are unachievable.

Methodology

Each robot’s body schema and peripersonal space are learned via self-organized map clustering. During run-time, action candidates are simulated in the forward model, scored by task reward. A shared planner sequences individual goals into cooperative behaviors. In an industrial assembly task, two robots share workspace to assemble fuse-boxes. We describe coupled forward models of body and peripersonal space learned via exploration. Internal simulation evaluates action-consequence pairs, driving cooperative sequences even when solo goals are unachievable.