HYBRID DELIBERATIVE/REACTIVE ARCHITECTURE FOR HUMAN-ROBOT INTERACTION

Abstract

In this paper we present a hybrid architecture that combines deliberative motion plan, local reactive behaviors and human-initiated inputs. The deliberative motion plan is described by an approximated navigational function created over a map of the environment. This function has a unique global minimum at a desired goal destination, and by using this function the robot can navigate autonomously to the goal. While navigating the robot senses the world and detects unpredicted obstacles along the way, and by using local reactive behaviors the robot avoids obstacles without disregarding the motion plan. At any point, the user can give inputs that are incorporated into the system. We combine these three distinct and at times conflicting control inputs in a proper way. We implemented this architecture on a robotic wheelchair and we show some experimental results. In our experiments the user uses a joystick to provide inputs to the system.