0 100 200 300 400 500 600
0
0.5
1
1.5
2
2.5
3
3.5
4
4.5
5
Useful Message Size in Bytes
Average Latency in ms
Effect of backchannel on Latency
Without backchannel on wireless network
With backchannel on wireless network
Fig. 6. Measured effect of backchannels on latency
Tests to determine how latency is affected by the addition
of a backchannel also match theoretical results. The tests
described do not measure only TCP latency because we timed
the round trip which includes time to parse the message.
The total savings in latency is an average of 20% on the
wireless network with our implementation, but can be as much
as 80% savings for messages in the 375 to 500 byte range, a
significant benefit for real-time systems.
VI. APPLICATION TO SEARCH AND RESCUE ROBOTICS
Urban Search and Rescue (USAR) involves coordination
between people, agents, and robots to explore a space with
many environmental challenges, including limited and spo-
radic communication. We have built a heterogeneous set of
wheeled differential drive robots, and several user interfaces
to allow robots to explore a mock up disaster site. In order
to enable robots, agents, and people to work together, we are
using the RETSINA multi-agent system architecture. Based on
empirical testing of our system, we found that the overhead of
KQML messages was very significant. Additionally, sending
real-time video to the human operators could only be done
at enormous cost through RETSINA and so was controlled
outside the architecture. After integrating backchannels into
the RETSINA architecture, we found that the stability and
reliability of the robot system improved significantly as latency
was decreased and throughput was increased.
VII. RELATED WORK
Discussion of the balance between efficiency and readability
is common in the networking community where HTML is
mixed with other formats using MIME headers to allow media
types other than simple ASCII text to be encoded into a
message. Much work has also been done to improve the speed
of transferring images and multimedia. This work [10], [11],
[14], [13] has contributed to the analysis on throughput and
is an excellent resource for more information on the effect
of message size on latency. To our knowledge, this is the first
attempt to integrate backchannels into a MAS in a formal way.
The MAS community has largely ignored the importance of
throughput efficiency. Work has been done to develop real-time
multi-agent systems, [15], [16], but this work has focused on
higher level algorithms, ignoring the low level protocol and
the potential benefits of improving throughput.
VIII. CONCLUSION
We have presented arguments that a single line and language
for agent communication is inadequate for systems that require
the transfer of multimedia files and low level data at high
frequencies. We describe a two-tiered communication archi-
tecture using backchannels and the steps necessary to integrate
backchannels into existing MAS architectures in a principled
way. We hope this flexibility will open up multi-agent systems
for use with physical agent systems, helping to bridge the
multi-agent multi-robot system research gap. Theoretical and
experimental tests show improved communication efficiency
and the addition of backchannels to the MAS RETSINA
enabled the use of RETSINA for an urban search and rescue
multi-robot system.
REFERENCES
[1] T. Finin, R. Fritzson, and R. McEntire, “KQML as an agent commu-
nication language,” in Proceedings of the 3rd International Conference
on Information and Knowledge Management, November 1994.
[2] (1997) Foundation for intelligent physical agents. [Online]. Available:
http://www.fipa.org
[3] K. Sycara, K. Decker, A. Pannu, M. Williamson, and D. Zeng, “Dis-
tributed intelligent agents,” IEEE Expert, December 1996.
[4] H. Chalupsky et al., “Electric elves: Agent technology for supporting
human organizations,” AI Magazine, Summer 2002.
[5] K. Decker, K. Sycara, and D. Zeng, “Designing a multi-agent portfolio
management system,” in Proceedings of the AAAI Workshop on Internet
Information Systems, 1996.
[6] O. Shehory, G. Sukthankar, and K. Sycara, “Agent aided aircraft
maintenance,” in Proceedings of Autonomous Agents ’99, May 1999,
pp. 306–312.
[7] K. Sycara, M. Paolucci, M. van Velsen, and J. Giampapa, “The
RETSINA MAS infrastructure,” special joint issue of Autonomous
Agents and MAS, vol. 7, no. 1 and 2, July 2003.
[8] C. Li, J. Giampapa, and K. Sycara, “A review of research literature on
bilateral negotiations,” Carnegie Mellon University, Robotics Institute,
Pittsburgh, PA, Tech. Rep. CMU-RI-TR-03-41, Nov. 2003.
[9] G. Zlotkin and J. Rosenschein, “Negotiation and task sharing among
autonomous agents in cooperative domains,” in Proceedings of the
Eleventh International Joint Conference on Artificial Intelligence, 1989,
pp. 912–917.
[10] N. Cardwell, S. Savage, and T. Anderson, “Modeling tcp latency,” in
Proceedings of IEEE INFOCOM, March 2000.
[11] M. Harchol-Balter. (2002, November) Quality of service lectures.
[Online]. Available: http://www-2.cs.cmu.edu/∼srini/15-441/F02/
[12] J. Wang, M. Lewis, and J. Gennari, “USAR: A game based simulation
for teleoperation,” in Proceedings of the 47th Annual Meeting of the
Human Factors and Ergonomics Society, Denver, CO, Oct. 13–17, 2003.
[13] J. Nagle, “Congestion control in IP/TCP internetworks,” Network Infor-
mation Center, SRI International, Menlo Park, CA, RFC 896, January
1984.
[14] J. F. Kurose and K. W. Ross, Computer Networking: A Top-Down
Approach Featuring the Internet. Addison-Wesley, July 2000, ch. 3.7
TCP Congestion Control.
[15] M. Allouche, O. Boisser, and C. Sayettat, “Temporal social reasoning
in dynamic multi-agent systems,” in Proceedings of the Fourth Inter-
national Conference on Multi-Agent Systems (ICMAS-2000). IEEE
Computer Society, 2000, pp. 23–28.
[16] V. Julian and V. Botti, “Developing real-time multi-agent systems,” in
Fourth Iberoamerican Workshop on Multi-Agent Systems, Iberagents
2002. IBERAMIA 2002, the VIII Iberoamerican Conference on
Artificial Intelligence, November 2002.