Godot, the Robot
Most research on spoken dialogue has focused on humans talking to virtual agents, often only reached at the end of a telephone line. Interesting challenges and opportunities arise when the interlocutor is a physically embodied mobile agent -- for example, a robot. When we enter into dialogue with a robot, we can talk about the physical environment that we share with the robot, and we get a palpable indicator of dialogue success when an utterance such as Go to the corridor! produces the desired effect. We have integrated our robot, Godot, with a spoken dialogue system, aiming to explore the new vista for human-computer interaction design that is involved with dialogue research with mobile robots. Godot's Specifications
Godot is an RWI Magellan Pro mobile robot platform with an on-board PC running Linux. It is cylindrical, about 50 cm high and 41 cm in diameter. Godot's sensor equipment consists of 16 sonars, 16 infrared sensors, 16 bumpers, an odometry component, and a Sony EVI-D30 colour video camera with a pan-tilt unit. The on-board computer is connected to the local network via a wireless LAN interface.
Godot's navigation system relies on sonars, infrared sensors and odometry. However, the camera is used as live feedback to the user, who can look ``through the eyes'' of Godot while engaging in a dialogue. Godot moves about in the basement of our department and uses an internal map for navigation. It has two levels of representation: a geometrical and a topological layer. Godot's Dialogue System
The dialogue system is implemented on top of OAA 2.1.0, a piece of middleware supporting a variety of programming languages. OAA agents can run on different machines and even on different platforms, and they communicate with each other by posing solvables via a facilitator, which is an example of a hub-architecture. The system contains OAA-agents for speech recognition (Nuance 8.0), speech synthesis (Festival), resolution (resolving natural language ambiguities, using Discourse Representation Theory as semantic formalism), inference (building models or finding counter-proofs), and dialogue management. To assist the inference agent, there are two further agents for model building (MACE) and theorem proving (SPASS). The system is coordinated by the dialogue manager, which triggers new OAA-solvables for speech recognition, speech synthesis and inference. It can also request information as the the robot's position and its current environment. This is made possible by allowing solvables as effects in update rules. Godot's Navigation System
The navigation system consists of three components which run concurrently on the on-board PC of Godot. They are Linux processes which communicate with each other via CORBA objects. Two of them, the map server and the navigation OAA-wrapper, are also OAA-agents. The map server stores Godot's internal representation of the environment as CORBA objects. Information stored in these objects can be retrieved by the navigation and dialogue modules via CORBA or OAA connections. The navigation module has a CORBA object to store information that is shares with the dialogue system. The OAA-wrapper has access to this object and OAA agents in the dialogue system can communicate with the navigation module via the wrapper. The dialogue system can monitor the current state of the navigation system, e.g. the current location of the robot in the map, and can send commands to it. In addition, there are modules for the camera view and for monitoring the location and orientation of the robot in the map. Videos of Godot
The following videos demonstrate a spoken dialogue with Godot. They show either the robot itself or a screenshot of the dialogue manager, where you can see the camera view of the robot and the discourse representations generated.
People currently working on Godot
Publications on Godot
Tetsushi Oka, 16 Dec 2004 |