A Framework in which Robots and Humans Help Each Other

Within the context of human/multi-robot teams, the “help me help you” paradigm offers different opportunities. A team of robots can help a human operator accomplish a goal, and a human operator can help a team of robots accomplish the same, or a different, goal. Two scenarios are examined here. First, a team of robots helps a human operator search a remote facility by recognizing objects of interest. Second, the human operator helps the robots improve their position (localization) information by providing quality control feedback. Introduction This paper reports on a preliminary investigation of collaboration in human/multi-robot teams. We have constructed a framework that is designed to support one or more robots working with a human operator in a dynamic, real-time environment. Control of the robots is shared between the human operator and a software controller, and the locus of control can switch during run-time. Sample studies are presented here to demonstrate the baseline capability of the framework and to drive the next step of our broader research plan. Our research is motivated by two related application areas: urban search and rescue (Murphy, Casper, and Micire 2001; Jacoff, Messina, and Evans 2000; Yanco et al. 2006) and humanitarian de-mining (Santana, Barata, and Correia 2007; Habib 2007). In both instances, teams of robots are deployed to explore terrain that is potentially unsafe for humans and to locate targets of interest. In the first case, robots explore an enclosed space, such as a collapsed building, and search for human victims who may be physically trapped. The goal is to locate these victims and transmit their positions to human operators, so that human first responders can remove the victims to safety. In the second case, robots explore an open space, such as a field in a war zone, to search for anti-personnel mines that may be hidden from view. The goal is to locate these mines and transmit their positions to human operators, so that the mines can be disarmed and the area rendered safe for people to traverse. Both application areas have three fundamental tasks in common. First, a robot must be able to explore a region (traverse and maneuver in the physical space) and localize (determine and track its position there). Second, a robot must Copyright c © 2011, Association for the Advancement of Artificial Intelligence (www.aaai.org). All rights reserved. be able to recognize objects of interest, using on-board sensors and possibly augmented intelligence to interpret sensor input. Third, a human operator must be able to communicate with the robots remotely and strategize so that the team can accomplish its overall task effectively. In a collaborative system, the human operator should not be overloaded with tasks, and the robots should not be idle; the robots should help the human operator accomplish her goal, and the human operator should help the robots accomplish their goal(s). Background and related work Human-Robot Interaction (HRI) supports collaborative activities by humans and robots to achieve shared goals. Typical HRI research concentrates on the development of software and/or hardware to facilitate a wide range of tasks. These include robots maneuvering in physical spaces, both those designed for humans (Kang et al. 2005) or unfit for humans (Murphy 2000); people programming complex robots (Sandini, Metta, and Vernon 2007); robots cooperating with human partners (Burke and Murphy 2004; Finzi and Orlandini 2005; Wegner and Anderson 2006). and with other robots (Dias et al. 2004; Lagoudakis et al. 2004; Mataric, Sukhatme, and Ostergaard 2003; Stone and Veloso 1998); and interfaces for communicating with robots (Kaber, Wright, and Sheik-Nainar 2006; Rooy, Ritter, and St Amant 2002). Deployed HRI applications include cleaning (Roomba 2010), helping the elderly (Tyrer et al. 2006), assisting first responders in search and rescue tasks (Crasar 2010), and de-mining in military settings (Freese et al. 2007). There are three main categories of control architectures for human-robot systems (Goodrich and Schultz 2007): fully autonomous, where robots make decisions and control their actions on their own; directly controlled, where robots are driven by human operators; and mixed-initiative (Carbonell 1971; Horvitz 1999), where robots share decision making with human users. Mixed-initiative systems reflect recent trends within the HRI community toward socially intelligent interfaces (Breazeal and Scassellati 2002; Dautenhahn 2007) in which the aim is for robots and humans to respond to each other naturally. We focus on mixed initiative architectures and highlight several approaches. Adjustable autonomy in a human-robot system permits dynamic transfer of control from human to 54 Help Me Help You: Bridging the Gaps in Human-Agent Collaboration — Papers from the AAAI 2011 Spring Symposium (SS-11-05)