19 Junho 2017, 17:40 - Fátima Sampaio
Candidate: Tiago Guiomar Ribeiro Nº 55446/D
Title: Creating the Illusion of Life in Autonomous Social Robots
Location: Sala 0.19, Pavilhão de Informática II, IST, Alameda
Advisor: Professora Ana Maria Severino de Almeida e Paiva
Abstract: Robots are becoming a new form of animated characters, and are being deployed into our society to be used in various social settings that can benefit of the use of technology and artificial intelligence (AI), such as education, entertainment or assisted living. This thesis explores how such social robots, through their physically expressive embodiment, and considering their autonomous capabilities, may be able to convey the illusion of life just as movie characters do, while interacting with humans. We establish and describe a new form of animation, called robot animation, which sets to bring the existing knowledge and techniques from traditional and CGI animation, into the field of autonomous social robots.
Various technical achievements were developed in order to support this quest. In particular, we present the SERA model and methodology for creating autonomous socially expressive robots, which relies in user-centred design and includes non-technical experts such as psychologists and character animators. An innovative animation engine called Nutty Tracks was created to support the blending, during interactions, of animations and postures pre-designed by artists, with motion that is procedurally generated in real-time. Nutty Tracks bridges the symbolic level of an autonomous artificial intelligence agent, with the lower-level of motion generation and control. This allows us to create autonomous social robots that are capable of conveying the illusion of life, in a way that users are also able to understand its communicative intentions. In order to support complex, articulated robots such as industrial manipulators, we have created ERIK, which is an expressive kinematics technique that acts by bring together inverse kinematics (IK) control with forward kinematics (FK) control. The blending of these two techniques provides a robot with the ability to express itself through posture while holding an orientation constraint towards a human or some part of the environment. It contrasts with current practices that allow to, in the best case, switch between IK and FK, but not to use both at the same time. However, in order for the solution to hold an orientation constraint, and to comply with the robot’s mechanical limits, the intended expressive posture suffers some distortion.
The Ahoy interactive scenario was developed in which humans play a game of pantomime with the custom-built Adelino robot. The robot could use its expressive posture to convey hints to the players, while keeping an orientation constraint towards their face. A user study was ran and showed that the users were still able to decode the intention of the robot even if the IK solution was distorting the pre-designed postures. The results provided initial evidence that expressive postures (controlled using FK) could be used along with IK in order to provide arbitrary robots with an animation model that works out-of-the-box, with nearly no tweaking. This contrasts with a data-driven approach which would require either a learning-by-demonstration process, or training from a pre-built corpus of postures, design specifically for a particular robot. In the long term, our technique is a basilar step towards autonomous expressive robots that exhibit the illusion of life, while interacting with humans and their surrounding environment.