Lecture 1: Introduction to Cognitive Robotics

Who is Serge Thill?

  • Cognitive science background (BSc), computational neuroscience (MSc), specialized in computational modeling
  • Fundamental interest in human cognition
  • Also interested in AI and intelligent technology as a tool for studying human cognition/implementing cognitive models.

Motivation of this course

Why cognition? Not necessary for staying alive. It's necessary for social interaction to communicate with other intelligent beings.

  • Understanding interaction requires understanding cognition
  • May lead to better insights on how machines can be sociable
  • But if may also lead to a better understanding of human cognition
    • They also force us to understand how humans interacts with machines

So robots are a method for studying cognition and social interactions. We focused on the former.

Simulator

Simulator: CoppeliaSim (free and open-source) for robot simulation.

Has a nice API and controllable from Python, Java, Rust (I think). We also going to use spiking NNs using Nengo.

Assessment

First half of the course: production of an opinion article.

  • Magazine style, e.g. Scientific American
  • Target: interested and educated layman
  • Relevance, clarity, strength of argumentation, clear position, good style, polished presentation, ability to remain on topic.

Second half: individual project work with report.

  • Simulated robot using CoppeliaSim
  • Report on 1) implemented agent, 2) reflection on possibilities and limitations in present-day cognitive robotics, given your experience with your own implementation.

W.r.t. the latter point, you will get some straightforward tasks, but also some tasks that are extremely challenging. These latter are supposed to show you the gaps in theory that need to be filled to efficiently implement ideas in practice. The second goal of the report is to reflect on these encountered issues.

Different definitions and ideas of goals for HRI

No longer a focus of this course this year.

HRI is a field dedicated to understanding, designing, and evaluating robotic systems for use by or with humans. Here the focus is on humans. The goal is to understand and shape human-robot interactions. (Goodrich & Schutlz, 2007).

In other definitions, the focus is more on design of algorithms and applying principles to robotics for good HRI.

More definitions in the slides.

The crux is that you can build robots to understand human cognition. You can also build robots to interact with humans. In both cases, you also somehow then need to learn about human cognition.

Favorite AI definition of Serge: AI is whatever a computer currently cannot do. Once we understand what the computer does (i.e. we understand the applied "rules") we won't call it AI anymore. The veil of mystery is gone.

Also: "Once we actually build system AI quickly degenerates into machine learning."

What is a robot?

See slides for extensive table and graphs. There is a spectrum for describing the types of robots that exist.

Should we care about human cognition?

Exhibit A: two examples of robots. There is nothing cognitive about their behavior. What's cognitively interesting is that we very easily attribute cognitive behavior to these robots.

Exhibit B: didn't pay attention.

Exhibit C: didn't pay attention.

(Sidenote: Most things you can perceive you can also change; otherwise perception would (evolutionary) be useless).

Key point w.r.t HRI

  1. We have to consider human cognition
    • When robots and humans interact, we need to take into account human behavior
    • Good interaction may require human-like human-understandable behavior
    • Interaction is relevant even outside of the strict scope of HRI (e.g. AVs).
  2. Human cognition is a good inspiration for building robots
  3. Models of human cognition might be usefully studied in robotic bodies

These routes lead to cognitive robotics!

Cognitive architectures

See slides. Book of David Vernon recommended. There are also very different types of cognitive architectures.