Travel agency dialogue task

Significance of travel agency dialogue task for a robot

In this competition, the participants will challenge a dialogue task at travel agencies. The robot as a counter salesperson responds to the customer's request through dialogue. Nowadays, many web services are available for looking for travel destinations and transportation methods. But still many people come to travel agencies. One of the reasons is that customers want to listen to the opinions and impressions of counter salespeople and find a place to travel that meets their needs. In this way, dialogue at a travel agency is not just about getting the information you need, it is a dialogue that requires hospitality. Therefore, a robot that can behave like a human can provide better dialogue compared to a voice-only dialogue system.

Travel destination recommendation task

Various dialogues are held at actual travel agencies, but in this competition, the participants will develop a robot that will engage in dialogues in the following situations. (This is called a "travel destination recommendation task".)

  • The robot acts as a counter salesperson.

  • The purpose of the customer is to decide on a sightseeing spot (one place) to enjoy the surrounding area of the travel agency. The customer decides two places as candidates before the dialogue and decides which one to go to in the dialogue with the robot.

  • The robot will explain the highlights of the two sightseeing spots, answer questions from the customer, and let the customer decide which sightseeing spot to visit.

  • The purpose of the robot is that the sightseeing spot that the agency recommends is chosen by the customer with satisfaction.

The situation of the task

  • The task for the robot is to recommend tourist spots around a certain facility.

  • In the preliminary round, the conversational robot of each team is placed in a shopping mall or at a public facility, where the people visiting these places talk to the robot and evaluate its performance by questionnaire. In the final round at the IROS venue, designated dialogue researchers as well as those working in the tourist industry talk to the robot for evaluation.

  • The robot and the customer face each other across the counter table as shown in the following figure (Android I on the right). The relative position of the robot, customer, and monitor is predetermined.

  • The robot talks with one customer in a dialogue.

  • One monitor is placed on the table so that the customer and the robot can see the same information at the same time (the position and orientation are fixed).

  • The customer and the robot talk in Japanese.

  • The customer fills out a questionnaire to answer the impression about the dialogue and robot.

The customer's behavior in the task

  • The customer is asked to follow the instructions in advance of the dialogue.

    • If you are planning to visit a shopping mall or museum on vacation and have the purpose of deciding on a sightseeing spot to go to for a day in the vicinity, please consult with the robot who is the counter salesperson.

    • Before having a dialogue, please select two sightseeing spots that you would like to visit from the candidate list (six spots around the shopping mall or museum). Please consult with the robot and select one of the two sightseeing spots that you would like to visit even if you pay for it.

    • You will have a dialogue with the robot for about 5 minutes.

    • Please check the information about the two sightseeing spots and decide the sightseeing spot you want to go to.

    • Sit in the counter chair before starting the dialogue.

    • After the dialogue is over, please select the spots you want to go to (you must select one).

    • Finally, please answer the questionnaire about your impression of the experience.

What to do with the robot in the task

  • Explain to the customer two candidate spots.

  • Interact with the customer through dialogue.

  • Aim to have the customer choose the sightseeing spot which the robot recommends as if it was chosen by the customer's will.

Regulation of the task

  • What the participants in the competition may let the robot do

    • Have the robot speak at any time.

    • Move the robot's gaze, facial expression, head, upper body, etc. at any time.

    • Use the sightseeing spot information data (given to the participant in advance) to explain the sightseeing spots and answer the customer's questions.

    • On the monitor on the table, the photos of the two spots are displayed. The robot may explain those photos displayed on the monitor.

    • If the customer asks for the robot's impressions and opinions about the spots, the participant may prepare the robot's impressions and opinions in advance and let it answer them. However, as a development guideline, the participant should develop a system that can respond even if other sightseeing spot information data is given in the competition.

    • The robot's speech synthesis shall use the specified speech synthesis system.

    • If the participants are required to use Amazon Polly, Japanese female voice needs to be used. The parameters of phonation and vocal-tract-length need to be set as follows. Any values can be used for the other parameters.
      <amazon:effect phonation="soft" vocal-tract-length="-15%">

  • Information that the robot can use during dialogue

    • Sightseeing spots information given to the participants in advance.

    • Information on the customer's position, voice, facial expression, gender / age (based on the face image) recognized using the camera and microphone installed around the robot.

    • Information on the position of the monitor (where it is placed).

  • How to handle sightseeing spot information

    • The sightseeing spots information database (in Japanese) is given in a CSV format.

    • The records in the database can be freely manipulated and used by the program.

    • In this competition, the participants will be given the information on sightseeing spots in the vicinity of the places where the preliminary or final rounds will be held, but as a development guideline, the participants should develop a system so that the robot can respond even if the sightseeing spot information is given in the competition which is not included in the given database.

  • The candidate and recommended sightseeing spots

    • The candidate sightseeing spots: Before starting the dialogue, the customer selects two sightseeing spots that she/he wants to visit from the six sightseeing spots (six spots are fixed through the competition) near the place where the preliminary or final round is held.

    • The recommended sightseeing spot: Randomly selected from the two spots selected by the customer.

  • What the participants should not let the robot do

    • Ask the customer to select the recommended spot by a method other than dialogue (e.g., a robot will give the customer a prize if the customer selects the recommended spot).

  • Start / end of the task

    • Before starting the task, the participant inputs the two candidates, the sightseeing spots (A, B) that the customer selected and the recommended sightseeing spot randomly selected into the participant's program. (Specifically, since the IDs of sightseeing spots (A, B) and the recommended spot are stored on the specified server, the information is received by the participant's program before the task starts. The ID is acquired in a list format. In the two candidate spot IDs, the first is A (displayed on the left side of the monitor) and the second is B (displayed on the right side of the monitor).

    • The photos of the two spots A and B are displayed side by side on the monitor of the table as shown in Figure 5 (A is on the left, B is on the right) (output to the monitor is performed by the organizer).

    • When the competition staff inputs the two spots heard from the customer into the server, the robot system starts counting the dialogue time and the staff guides the customer to sit in the chair.

    • Start the participant's program when the customer seats in a chair.

    • After 5 minutes and 30 seconds have passed since the start of the count (30 seconds is a margin for the case where the start process takes time), the participant's program will behave as if the robot ends the dialogue (the robot is expected to say, "I'm sorry, but it's time, so I'd like to end the guidance.", etc.) and end the dialogue.

    • It doesn't matter whether the robot initiates the dialogue or waits for the customer to start talking.

  • Guidelines for asking the customer's personal information (name, age, etc.) during the dialogue

    • The robot may ask the customer for information necessary for planning the trip, such as name, age, and preferences.

    • The customer is instructed as follows before the task.

      • "Please note that the robot may ask you for personal information such as your name and age that you need to plan your trip. If you don't want to say something personal, you don't have to answer, or you can answer with a pseudonym."

    • The participant should assume there will be a case that the customer may not answer when listening to the personal information in the dialogue.

How to evaluate the robot

The task achievement level is evaluated from the following two points of view.

  • Satisfaction of the customer (evaluation by questionnaire after dialogue).

  • Whether the customer choses the recommended sightseeing spot.

The participants with high rankings in both scores are highly evaluated. From the preliminary rounds, about 30 - 40 % of participants will advance to the final round. In the final round, the following items will be evaluated in addition.

    • Designated dialogue researchers and those working in the tourist industry who can evaluate the behavior of customer service (experts of counter salesperson) will be the judges. The judges will try a dialogue with the robot controlled by the participant program. The researchers and experts evaluate from different evaluation perspectives.