Introduction
In the field of social robotics, interaction has traditionally relied on visual and auditory cues, such as facial recognition and voice tone analysis. However, these modalities often fail to capture the intimacy and nuance of "affective touch," a fundamental component of human-animal and human-human bonding. Researchers at ETRI have developed KOBIE, a koala-like robot designed specifically to bridge this gap. KOBIE utilizes a sophisticated internal "Emotion Engine" and a high-density tactile sensor array to perceive and respond to physical stimuli like stroking, poking, and hugging with lifelike emotional transitions.
System Architecture: The Emotion Engine
The core of KOBIE’s intelligence is its Emotion Engine, which moves beyond discrete emotional states (e.g., jumping instantly from happy to sad) toward a continuous, multi-layered psychological framework. This engine is divided into three primary modules that process sensory input into behavioral output.
- Emotion Feature Generation Module (EFGM): This module serves as the sensory interface, transforming raw data from the robot's tactile sensors into distinct "emotion features."
- Internal Status Generation Module (ISGM): The ISGM acts as the robot's cognitive core. It integrates a Needs Model based on Maslow’s Hierarchy, prioritizing physiological requirements (like rest) and safety before social needs. It also maintains a Mood Model (tracking six parameters such as comfort and irritability) and a 3D Emotion Model. This 3D model uses a vector space defined by Pleasantness ($E_p$), Activation ($E_a$), and Certainty ($E_c$), allowing for smooth, organic transitions between states.
- Behavior Decision Module (BDM): This final stage translates the internal emotional vector into specific physical actions, sounds, or facial expressions selected from a library of 22 distinct behaviors.
Hardware and Tactile Sensing
To facilitate natural interaction, KOBIE is designed with 21 Degrees of Freedom (DOFs), with a significant concentration in the head (5 DOFs) to enable expressive movements of the eyelids, ears, and neck. The robot's "skin" is embedded with a variety of sensors to detect the location and intensity of touch.
As shown in the sensor map, KOBIE utilizes Force Sensing Resistors (FSRs) on the head, back, sides, and chin to measure pressure, while binary on/off sensors on the legs, paws, and tail detect simpler contacts. This configuration allows the robot to distinguish between different types of affective stimuli, such as a gentle stroke versus a sudden hit. These inputs are mapped to an "Emotion Factor Table," where positive interactions like stroking contribute to "Happy" factors, while internal states like fatigue contribute to "Unhappy" factors.
Key Insights and Behavioral Mapping
A critical innovation in KOBIE is the integration of internal "drives." By utilizing Maslow’s hierarchy, the robot does not merely react to the user; its response is modulated by its own internal state. For instance, if the robot is in a state of high "fatigue," it may respond listlessly to a stimulus that would otherwise elicit a playful reaction. This dependency on internal needs makes the robot's behavior feel less like a programmed machine and more like a living companion.
The system successfully maps these complex internal states to a behavioral library that includes "cuteTricks," ear-wagging, and specific vocalizations for anger or shame, ensuring that the user receives clear, emotionally resonant feedback from their physical interactions.
Conclusion and Future Work
KOBIE represents a significant step toward more naturalistic human-robot interaction by prioritizing tactile affect and continuous emotional modeling. While the current architecture excels at real-time reactivity, future iterations will focus on two key areas: a Memory Model to allow the robot to recognize and build long-term relationships with specific users, and a Personality Model to ensure that different units exhibit unique behavioral traits. These additions aim to further deepen the social bond between humans and their robotic companions.
Summary updated