The Six Hug Commandments: Design and Evaluation of a Human-Sized Hugging Robot with Visual and Haptic Perception

1. Problem Statement

While hugging offers significant social and physical health benefits (e.g., stress reduction, oxytocin release), the lack of social touch—exacerbated by remote interactions and the COVID-19 pandemic—negatively impacts well-being. Existing robotic solutions for hugging suffer from critical limitations:

• Lack of Autonomy: Many require teleoperation (a remote human operator) or manual triggering (button presses), preventing spontaneous interaction.
• Poor Adaptability: Most robots cannot adapt to the user's body size, shape, or position, leading to awkward or unsafe embraces.
• Safety and Comfort Issues: Previous iterations (e.g., the PR2-based HuggieBot 1.0) were often too large, rigid, or required unnatural user actions (e.g., pressing a sensor on the robot's back) to initiate or end a hug.
• Absence of Closed-Loop Perception: No prior system successfully integrated visual perception for initiation and haptic sensing for sizing and release detection in a fully autonomous, human-sized platform.

2. Methodology

A. Theoretical Framework: The Six Hug Commandments

The authors propose six tenets ("commandments") to guide the design of natural and enjoyable robotic hugging. The first two were previously validated; the latter four are new contributions of this work:

1. Soft: The robot must have a soft exterior.
2. Warm: The robot must provide thermal warmth.
3. Human-Sized: The robot should be sized similarly to an adult human.
4. Visual Perception: The robot must visually detect an approaching user and initiate a hug autonomously (no button press).
5. Haptic Sizing: The robot must automatically adjust its embrace to the user's specific size and position.
6. Reliable Release: The robot must reliably detect and react to the user's desire to end the hug, regardless of arm position.

B. System Design: HuggieBot 2.0

To implement these tenets, the authors built HuggieBot 2.0, a novel humanoid robot featuring:

• Frame & Arms: A custom stainless steel V-shaped base for stability. It utilizes two 6-DOF Kinova JACO arms equipped with torque sensors at every joint. The arms are covered in soft foam and a sweatshirt.
• HuggieChest (Inflatable Torso): A custom inflatable PVC torso acting as a large-scale haptic sensor. It contains a pressure sensor (BME680) and a microphone (MAX4466) in the rear chamber to detect contact pressure and ambient noise.
• Thermal System: Heating pads (Thermophore MaxHeat) are integrated beneath the outer clothing to provide warmth.
• Visual Perception: An Intel Realsense RGB-D camera runs a deep-learning-based person detection model (TensorFlow SSD MobileNet) to identify approaching users. It uses depth data to confirm approach and initiates a hug when the user enters a 2.45m threshold (social to personal space transition).
• Control Logic:
  • Initiation: Visual detection triggers a verbal request ("Can I have a hug, please?") and arm movement.
  • Sizing: Arms close until joint torque exceeds a threshold (10 Nm for shoulder, 5 Nm for elbow), adapting to the user's body size.
  • Release: Detection of user release occurs via two methods: a drop in chest pressure (indicating separation) or a high torque spike (>20 Nm) on the arm joints (indicating the user is pushing away).

C. Experimental Evaluation

The study involved two phases:

1. Online User Study (N=117): Participants watched videos comparing HuggieBot 2.0 against the previous PR2-based HuggieBot 1.0. They rated embodiment, movement, and preference.
2. In-Person User Study (N=32): Participants exchanged 8 hugs with HuggieBot 2.0 under all combinations of three binary factors:
   • Visual Initiation (Yes/No)
   • Haptic Sizing (Yes/No)
   • Haptic Release Detection (Yes/No)
   • Note: A 8x8 Latin square design was used to counterbalance order effects. Users rated safety, naturalness, enjoyability, intelligence, and friendliness after each hug.

3. Key Contributions

• The Six Hug Commandments: A formalized set of design principles for social-physical HRI, validating four new tenets (Human-sized, Visual, Adaptive Sizing, Reliable Release).
• HuggieBot 2.0 Platform: The first human-sized hugging robot to integrate visual (for initiation) and haptic (for sizing and release) perception in a closed-loop system.
• Novel Inflatable Sensor (HuggieChest): A soft, inflatable torso with integrated pressure and audio sensing to detect contact and release without rigid components.
• Empirical Validation: Comprehensive data demonstrating that haptic reactivity (sizing and release) significantly improves user perception, while visual initiation shows positive (though statistically non-significant in this specific setup) trends.

4. Results

Online Study Findings

• Preference: Users significantly preferred HuggieBot 2.0 over HuggieBot 1.0 (p < 0.001).
• Reasons: The PR2 (HuggieBot 1.0) was perceived as "over-powering" and "unnatural" due to its size and the requirement for users to press a button on its back. HuggieBot 2.0 was rated as "friendlier" and more intuitive.
• Appearance: Users preferred the robot's soft, clothed appearance (sweatshirt over a purple robe) over the exposed mechanical look.

In-Person Study Findings

• Haptic Sizing (Tenet 5): This had the most significant impact.
  • Significantly increased perceived naturalness (p < 0.001).
  • Significantly increased enjoyability (p < 0.001).
  • Significantly increased perceived intelligence (p < 0.001).
  • Significantly increased friendliness (p < 0.001).
• Haptic Release (Tenet 6): Users strongly preferred haptic release over timed release. 53% voiced a preference for haptic release, noting that timed releases felt "too short" or that the robot "didn't want to hug me."
• Visual Initiation (Tenet 4): While 25% of users verbally preferred not pushing a button, the statistical analysis did not show a significant difference in ratings between visual and button-initiated hugs. The authors attribute this to the robot's slow arm speed causing awkward waiting times, which may have masked the benefits of visual initiation.
• Warmth (Tenet 2): Physical warmth was mentioned positively 39% of the time, confirming its critical role in pleasant interactions.
• General Attitude: After the experiment, users reported feeling more understood by the robot (p = 0.0025) and having increased trust (p < 0.001).

5. Significance and Conclusion

This research establishes that haptic reactivity is the most critical factor in making a hugging robot feel natural, intelligent, and friendly. While visual perception is desirable, its current implementation was limited by mechanical speed constraints.

The study concludes that adhering to the Six Hug Commandments significantly improves user perception of robots. Specifically, a robot that is soft, warm, human-sized, and capable of adaptive, closed-loop hugging (sensing size and release) transforms the interaction from a mechanical task into a socially supportive experience. This is particularly significant for providing social touch benefits during periods of physical distancing, offering a viable technological solution to loneliness and the lack of physical contact.

Limitations: The study noted that the slow speed of the Kinova JACO arms created awkward delays, particularly during visual initiation. Additionally, the self-selection bias of participants (who were already interested in robots) may limit the generalizability of the "fear" or "rejection" metrics, though it ensured a high success rate for the physical interaction.