Human-Robot Collaboration for Remote Surveillance

Evan Sultanik, Ilya Braude, Peter Thai, Robert N. Lass, Duc N. Nguyen, Joseph B. Kopena, William C. Regli, Sean A. Lisse, Steven N. Furtwangler, Alan J. VaydaAAAI 2008

Introduction

Traditional robotic surveillance has long been hampered by a significant manpower bottleneck. In conventional setups, managing a single robot often requires two or three human operators to handle movement, sensor data, and security simultaneously. This tele-operation model is not only labor-intensive but also mentally taxing, requiring constant high-bandwidth video feedback and leaving little room for high-level decision-making.

To address this "cognitive overload," researchers have proposed a shift toward a collaborative, semi-autonomous model. By increasing robot autonomy, the goal is to transform robots from mere remote-controlled tools into capable teammates that can handle low-level tasks like navigation and obstacle avoidance independently.

Technical Framework

The proposed system is built on a multi-agent system (MAS) architecture where each robot is controlled by a software agent. These agents utilize the Soar cognitive architecture to model reasoning processes and make decisions based on coordinated plans.

Communication is facilitated through a Mobile Ad-hoc Network (MANET) using the Optimized Link State Routing (OLSR) protocol. This infrastructure-less networking approach is critical for surveillance in remote or damaged areas. The system integrates hardware such as PDAs and tablets for human operators, allowing them to interact with the robotic agents via an Intelligent Control Framework (ICF) that supports task management and "dialog" between humans and machines.

The Robot as a Teammate

A central tenet of this research is treating the robot as a team member rather than a vehicle to be driven. This paradigm shift enables several key behaviors:

  • Interruptible Tasking: Robots can pause routine patrols to investigate new leads—such as a suspicious vehicle—and autonomously resume their original tasks afterward.
  • Shared Situational Awareness: Human operators can annotate digital maps on their handheld devices. These annotations are instantly propagated across the network, triggering automated responses from robotic agents.
  • Graceful Degradation: Local agents are designed to continue operating "sub-optimally" even if connectivity to a central command center is lost, ensuring mission continuity.
Figure 1: A field demonstration showing human operators and a robotic agent working as a collaborative team.

Experimental Results

The system was validated through a qualitative demonstration of a collaborative surveillance scenario. Key outcomes included:

  1. Reduced Oversight: Robots successfully maintained "follow-the-leader" and waypoint navigation behaviors without continuous human steering.
  2. Autonomous Investigation: In a highlight of the trial, a human operator identified a "suspicious vehicle" on a map. This event triggered a robotic agent to autonomously request permission to investigate. Upon approval, the robot executed the investigation loop independently.
  3. Network Resilience: The researchers successfully bridged local MANET traffic over a CDMA cellular network to a remote command center, demonstrating the feasibility of long-distance oversight for local autonomous teams.
Figure 4: The system interface showing a robot requesting permission from a human operator to perform a specific investigative task.

Limitations and Future Directions

While the demonstration was successful, several challenges remain. The authors noted that robot reasoning can become sub-optimal when disconnected from central command, indicating a need for more robust local intelligence. Furthermore, navigating unpredictable environments entirely autonomously requires more advanced sensory capabilities than those currently implemented. Future work will also need to address how these MANET-based agent systems scale when dozens of robots and humans are involved in a single operation.

Summary updated

BibTeX
@inproceedings{sultanik2008human,
  author = {Evan Sultanik and Ilya Braude and Peter Thai and Robert N. Lass and Duc N. Nguyen and Joseph B. Kopena and William C. Regli and Sean A. Lisse and Steven N. Furtwangler and Alan J. Vayda},
  title = {{Human-Robot Collaboration for Remote Surveillance}},
  booktitle = {AAAI Conference on Artificial Intelligence},
  year = {2008},
}
Figure