Family Robotics - Deep Q-learning with Samu

We believe that Samu will be the ancestor of developmental robotics chatter bots that will be able to chat in natural language like humans do.

Samu's Family Tree

  1. Samu The main purpose of this project is to allow the evaluation and verification of the results of the paper entitled "A disembodied developmental robotic agent called Samu Bátfai". It is our hope that Samu will be the ancestor of developmental robotics chatter bots that will be able to chat in natural language like humans do.
  2. Isaac "The son of Samu is Isaac." The project called Isaac is a case study of using deep Q learning with neural networks for predicting the next sentence of a conversation.
  3. Jacob "The son of Isaac is Jacob." The project called Jacob is an experiment to replace Isaac's (GUI based) visual imagination with a character console.
  4. Judah "The son of Jacob is Judah." In the project called Judah we equip Jacob with a text-based user interface.
    samu_tui
  5. Super-Judah
  6. Perez "The son of Judah is Perez." The project called Perez allows to perform experiments to test several parameter settings. Here each experiment is implemented as a separate git branch.
  7. Hezron "The son of Perez is Hezron." The project called Hezron gives the opportunity to experiment with different arrangements of SPO triplets in the visual imagery. hezron_tui
  8. Ram "The son of Hezron is Ram." The project called Ram allows to experiment with different deep Q learning algorithms to express feelings when he is talking. hezrontui2

The resources used in the paper entitled 'A disembodied developmental robotic agent called Samu Bátfai'

Experiments: JUDAH-A

Experiment A1

Experiment A2

JUDAH-B

The man's best friend is the car

The goal of the development of "The man's best friend"-type systems is to equip robot cars with a Human-Level AI interface. For background information, see the article Turing's Imitation Game has been Improved.

Resources for my Students

For more detailed information, see the Research and Education Plan (it is available in Hungarian only).

Competition Tasks

In the paper 'A disembodied developmental robotic agent called Samu Bátfai', we did not focus on the issue of triplet extraction. But certainly, it is a very important data mining component of Samu and, of course, it is not an easy task. Therefore we are going to try to improve it in programming competitions. The object of the first competition is to see who can extract the most number of triplets from the Gospel of Matthew of the BBE (Bible in Basic English)

Samu

One of the purpose of the original Samu project was only to allow the evaluation and verification of the results of the paper entitled "A disembodied developmental robotic agent called Samu Bátfai". This paper presents Q learning with neural networks approximators used by Samu. It is our hope and goal that Samu will be the ancestor of developmental robotics chatter bots that will be able to talk and read in natural language like humans do. At this moment Samu is the generic name of rapid prototypes that facilitate achieving this goal.

Highlights of Samu

Samu

  1. directly implements the definition of machine consciousness (such as the definitions of conscious and intuitive computer programs) introduced in the paper entitled Conscious Machines and Consciousness Oriented Programming.
  2. corresponds the principles of Developmental Robotics, Some Basic Principles of Developmental Robotics.
  3. applies Q learning with neural networks approximators
  4. uses multilayer perceptrons for approximation of the Q function, it is a deep Q learning feature, Human-level control through deep reinforcement learning

Last modified: 23 Oct 2015

Norbert Bátfai, PhD
batfai.norbert@inf.unideb.hu
IT Dept, UD