Human-Computer Interaction for CSAI

by Albert Major

Introduction to Human-Robot Interaction (HRI)

Definition of Robotics

• Core Goal: Robotics is a research field dedicated to creating an intelligent connection between Perception and Action.
• Sensors vs. Perception:
  • Sensors: The acquisition of raw information (e.g., pixel data from a camera).
  • Perception: Processing that raw information into meaningful data (e.g., detecting a specific face within the camera feed).
• The 5 Essential Components of a Robot: To be classified as a robot, an agent generally needs these five features:
  1. Body: A physical presence (or simulated physical presence) in the environment.
  2. Sensors: Tools to gather data (cameras, distance sensors, touch sensors).
  3. Actuators: Components that allow movement or action (mostly motors).
  4. Controller: The "brain" that processes sensory input and generates action commands.
  5. Function: A specific task to perform in the environment (e.g., manipulation, locomotion, interaction).

Robots vs. Automation (The Coffee Machine Argument)

• Is a coffee machine a robot? Generally, no. While it has sensors (temperature) and actuators (pumps), it lacks an intelligent connection between perception and action.
• The Distinction: A coffee machine performs the exact same output (standard coffee) regardless of who is standing in front of it.
• Intelligent Agent: To be a robot, it would need to perceive the specific user (e.g., recognize a face) and alter its action based on that perception (e.g., "I see you prefer coffee at 75 degrees, not 80"). If a machine repeats the same task without adapting to the environment, it is automation, not intelligent robotics.

Human-Robot Interaction (HRI) vs. Pure Robotics

• Pure Robotics: Focuses on manipulating the environment to execute a task (e.g., a factory arm assembling a car). Efficiency is the priority.
• HRI: Focuses on robots that work with or around humans. This requires adherence to social norms (e.g., maintaining personal space, politeness, acknowledging presence).
• Key HRI Requirements:
  • Emotional Intelligence: Displaying or understanding emotions.
  • Embodiment: Having a physical body is crucial for social presence.
  • Productivity/Helpfulness: The interaction must aid the human or the task.

Challenges in Deployment (Horror Stories)

• Robots often fail when moving from the lab to the "wild" (real world).
• Mall Robot (Japan): Children became aggressive and abusive toward a robot in a shopping mall, disrupting its function.
• HitchBot: A hitchhiking robot successfully traveled Canada and Europe but was destroyed/vandalized shortly after entering the USA.
• Takeaway: Roboticists must design for extreme human behaviors, not just ideal scenarios.

Robot Taxonomy and Form

• Biomorphic: Resembles living biological forms in general.
• Zoomorphic: Specifically resembles an animal (e.g., a robotic dog or seal).
• Anthropomorphic: Resembles a human structure (e.g., Asimo, Pepper).
• Mechanoid/Abstract: Form does not resemble living things (e.g., Roomba, factory arms).
• The Uncanny Valley: As robots become more human-like, they become more appealing up to a point. If they become almost human but have slight imperfections (e.g., dead eyes, unnatural movement), they provoke a strong sensation of revulsion or creepiness.

Design Heuristics for HRI (Exam Topic) There are four main heuristics derived from design principles that robots should follow:

1. Form Follows Function: The physical shape of the robot should communicate what it does (e.g., big wheels suggest outdoor terrain).
2. Function Follows Form: If a robot looks like a specific thing, it must act like it (e.g., if it looks like a human, people expect it to have human-level intelligence).
3. Do Not Mix Metaphors: Do not combine conflicting biological signals.
   • Example of Violation: Professor James Spiegel designed a "household family member" robot that was a humanoid but had octopus tentacles.
4. Match Interaction with Form/Function: The way interaction occurs must align with the robot's capabilities.
   • Violation Example: The "Octopus Humanoid" violated almost all principles. It looked human but had tentacles (mixed metaphor), promised to be a family member but could not perform basic chores (over-promised, under-delivered).

Types of Interaction You are responsible for knowing these three main types:

1. Non-Verbal Interaction:
   • Often more dominant and truthful than verbal communication.
   • Context Example: If you ask someone "Did you install a spyware backdoor?" and they say "No" but nod their head "Yes," the non-verbal nod is the true signal.
   • Gaze: Eye direction indicates attention. It is difficult for robots to coordinate head and eye movements naturally.
   • Creative Cues: Robots can use cartoon-physics cues (e.g., sweating animation, smoke for anger, shape-changing) to express status.
   • Joint Attention: Both the robot and human looking at the same object creates a shared reality.
2. Verbal Interaction:
   • Uses speech and language.
   • Historically difficult because robots struggle with context and shared reality (knowing what "that thing over there" refers to).
   • Large Language Models (LLMs) are currently improving this capability significantly.
3. Spatial Interaction (Proxemics):
   • The distance and angle between agents carry meaning.
   • Face-to-Face: Suggests competition or direct instruction/focus.
   • Side-by-Side: Suggests collaboration or shared attention toward a task.
   • Robots must respect social distances (not getting too close to strangers).

Service Robotics

• Service robots are usually deployed for tasks that are Dirty, Dull (Repetitive), or Dangerous.
• Success Story: The Roomba (vacuum) is the most successful service robot because it functions regardless of the room layout (robustness).
• Healthcare/Therapy:
  • Paro (Seal robot) / Flat Cat: Used for touch-based therapy.
  • Keepon: Used for interacting with children, specifically those with autism, to help them recognize emotions.

Ethical and Practical Issues

• Addiction: Users (especially children) might form unhealthy emotional dependencies on robots.
• E-Waste: If a robot fails to be useful (teleoperated only, or over-promised utility), it becomes expensive trash.
• Interfaces: Current software interfaces for controlling research robots are often poor, designed by engineers for engineers, and lack usability standards found in web/app design.