The Fourth International Automated Negotiating Agents Competition

To be held in conjunction with the Nineth International Joint Conference on Autonomous Agents and Multi-Agent Systems (AAMAS 2014), Paris, France, 5th-9th May 2014.


ANAC 2014 Special Session at ACAN (Tuesday May 6, 17:10 - 19:00, Room: Ella Fitzgerald A)
ANAC 2014 Session at AAMAS (Thursday May 8, 10:30 - 12:30, Room: Le parc Montsouris, level 3)

Join us!

ANAC 2014 Special Session at ACAN (Tuesday May 6, 17:10 - 19:00, Room: Ella Fitzgerald A)
ANAC 2014 Session at AAMAS (Thursday May 8, 10:30 - 12:30, Room: Le parc Montsouris, level 3)

Overview of ANAC2014

The ANAC competition brings together researchers from the negotiation community and provides a unique benchmark for evaluating practical negotiation strategies in multi-issue domains. The four previous competitions have spawned novel research in AI in the field of autonomous agent design which are available to the wider research community. The focus of this year's competition is on nonlinear utility functions. The goals of the competition are

  • to encourage the design of practical negotiation agents that can proficiently negotiate against unknown opponents and in a variety of circumstances,
  • to provide a benchmark for objectively evaluating different negotiation strategies,
  • to explore different learning and adaptation strategies and opponent models, and
  • to collect state-of-the-art negotiating agents and negotiation scenarios, and making them available to the wider research community.


The aim for the entrants to the competition is to develop an autonomous negotiation agent. Performance of the agents will be evaluated in a tournament setting, where each agent is matched with all other submitted agents, and each pair of agents will negotiate in a number of nonlinear negotiation scenarios. Negotiations are repeated several times to obtain statistically significant results. The winning agent will be the one with the highest overall score.

A negotiation scenario consists of a specification of the objectives and issues to be resolved by means of negotiation. This includes the preferences of both negotiating parties about the possible agreements. The preferences of a party are modelled using nonlinear, multi-issue utility functions.


Rules of Encounter

Negotiations are bilateral and based on the alternating-offers protocol. Offers are exchanged in real time with a deadline after 3 minutes. This means that the number of offers exchanged within a certain time period varies and depends on the computation required by the agents. If no agreement is reached by the deadline, or if either agent chooses to terminate the negotiation before the deadline, both agents receive their utility of conflict. In addition, there will be a discount factor in about half of the domains, where the value of an agreement decreases over time. The challenge for an agent is to negotiate without any knowledge of the opponent's preferences and strategy. Although each agent participates in many negotiation sessions, against different opponents, and in a wide variety of negotiation scenarios, agents cannot learn between negotiations. This means that negotiation agents only have the opportunity to adapt and learn from the offers they receive within a single negotiation session.

Agents can be disqualified for violating the spirit of fair play. The competition rules allow multiple entries from a single institution, but require each agent to be developed independently. Furthermore it is prohibited to design an agent which benefits some other specific agent. In particular, the following behaviors are strictly prohibited:

  • Designing an agent in such a way that it benefits some specific other agent.
  • Communicating with the agent during the competition.
  • Altering the agent during the competition.

The participants can use up to 2 GBytes of memory of their agent, if they use beyond that amount and the system cannot cope, their agent will be taken out of the competition.


The negotiation tournament was run using the java-based GENIUS negotiation platform, which has been developed to facilitate research in the area of bilateral multi-issue negotiation. It has an open architecture that allows for easy development and integration of existing negotiating agents using design patterns. GENIUS can be used to simulate individual negotiation sessions as well as tournaments between negotiating agents in various negotiation scenarios. The core functionality of the system includes:

  1. specification of negotiation domains and preference profiles;
  2. simulation of a bilateral negotiation between agents;
  3. analysis of the negotiation outcomes and negotiation dynamics.

It furthermore allows the specification of negotiation domains and preference profiles by means of a graphical user interface.

The GENIUS platform, together with the agents and scenarios from the previous competitions are available for download. More information about the platform can be found at the GENIUS web page. The agents from the previous competitions are available.

Main updates with respect to ANAC 2013

This year we extend the utility model to "nonlinear utility functions". Another challenge is to deal with large-size domains, with outcome spaces as big as 5010 outcomes.

Qualifying Round and Finals:

There will be initial qualifying rounds, and the top 8 performing agents will continue to the finals, which will be held at the AAMAS conference.

It is expected that teams that make it through to the finals will have a representative attending the AAMAS 2014 conference. Each team in the final will have the opportunity to give a brief presentation describing their agent.

Important Dates:

Deadline for uploading(submission) : 1 April, 2014



The prize money will be at least $1000. This will be shared among the top agents in two categories: (1) winners in terms of individual utility, and (2) winners in terms of social welfare (measured by obtaining the highest average product of utilities of both parties).


Questions and Answers

Feel free to ask your questions!

(Q.1) Do you intend to provide a minimal (non-linear) example domain?

(A.1) There are three non-linear scenarios that are currently distributed with the latest version of Genius, version 5.1. You can test your agent on these scenarios.

(Q.2) Do participants need to submit a domain along with their agent?

(A.2) No: participants of ANAC 2014 do not have to create a domain. The negotiation scenarios for ANAC 2014 will be generated by the ANAC organization and will be similar to the three scenarios that are included in Genius 5.1.

(Q.3) Should the SortedOutcomeUtilitySpace class not be used this time? As you are targeting 10^50 space sizes, I assume it wont be sortable efficiently.

(A.3) Since there will again be a 3 minute deadline in ANAC 2014, it is not feasible to sort the outcome spaces of the non-linear scenarios. SortedOutcomeUtilitySpace and other such classes cannot be used in ANAC 2014.

(Q.4) Will the getMaxUtilityBid() and getMinUtilityBid() methods of UtilitySpace class be available like before?

(A.4) These methods are available, but again, it is not feasible to use these methods in the non-linear scenarios. You will have to design a more appropriate search method yourself.

(Q.5) Will the getUtility() method return the (non-linear) utility value now?

(A.5) Yes: the getUtility() method will give you the utility of any offer, also in the non-linear scenarios. Since you get no information about the structure of the non-linear utility functions, you need to use this method to sample the utilities in the non-linear scenarios to search through the outcome space.

(Q.6) Are all domains in the competition going to be non-linear or will there be linear domains as well?

(A.6) The agent programs must be able to work on both of linear and nonlinear domains safely (without any computational failure). However, in the competition, your agent will be scored based on non-linear domains only.

(Q.7): What is the maximum number of agents each team is allowed to upload?
(A.7): Each team can only participate with one agent, but there can be multiple submissions per research institute, as long as the teams do not collaborate.

(Q.8): I can't find anything about the nonlinear evaluation function that is going to be used to calculate the utility of a bid. Is there anywhere I can find the utility calculation formula, since it is important to know how the utility is calculated especially when designing an opponent model? What I want to know is what is the exact formula that is used in nonlinear scenarios instead of this linear additive function.
(A.8): In the non-linear scenarios, the agents no longer have linear utility functions; instead, they can only sample their utility of a bid through the getUtility() method. The utility of an offer can be completely arbitrary, so there is no underlying structure you can use. In terms of the API of Genius, this means the agents no longer have access to methods pertaining to linear scenarios (e.g., getWeight()). Please use the getUtility() method to sample the utilities in the non-linear scenarios to search through the outcome space.

(Q.9) Is my agent allowed to do the searching for good offers in a separate thread?

(A.9) It is not allowed to start any new threads in your agent, so you cannot continue searching while it's the opponent's turn.





The ANAC2014 will be held at the Paris Marriott Rive Gauche Hotel & Conference Center
17 Boulevard Saint Jacques
Paris, 75014, France
33 1-4078 7980

Large Map - OpenStreetMap

Organising Committee

  • Dr. Reyhan Aydogan, Delft University of Technology
  • Tim Baarslag, Delft University of Technology
  • Prof. Katsuhide Fujita, Tokyo University of Agriculture and Technology
  • Dr. Koen Hindriks, Delft University of Technology
  • Prof. Dr. Takayuki Ito, Nagoya Institute of Technology
  • Prof. Dr. Catholijn Jonker, Delft University of Technology




€750 Xelvin, enterprising engineers,


€250 Makoto Lab., Inc.



For any questions, the main contact is
For technical questions contact Reyhan Aydogan <> and Katsuhide Fujita <>.