Connectionists: Event on "REAL 2021-2022 - Robot open-Ended Learning competition", 18/01/2022 15:00-16:30 (CET)

[Gianluca Baldassarre]

*** Event on the "REAL 2021-2022 - Robot open-Ended Learning competition"
(free participation) ***

On 18 January 2022, there will be an online presentation of the competition
followed by a hands-on demonstration. The free event will have this
organisation (times refers to CET - Central European Time):
*15:00-15:45*:
- objectives and scope of the competition;
- tour of the simulation: rules, organisation.
*15:45-16:30 *(and longer if needed or requested):
- installation of the competition software 'starting kit' on your computer;
- run of the 'baseline model' as a first solution to the competition
challenge;
- modifications of the baseline model;
- demo on how to develop new algorithms to test.

The event will be openly held in streaming with GoogleMeet at:
meet.google.com/qtj-wexv-pxo

** *Features of the REAL 2021-2022 - Robot open-Ended Learning competition*
**

The third edition of the REAL 2021-2022 competition ("Robot open-Ended
Autonomous Learning"), which started in 2021 and will end in 2022, aims to
develop a benchmark in the field of open-ended learning robots. Moreover,
the competition aims to form a community focused on comparing models able
to face the several interesting challenges posed by open-ended learning.

In the competition, a simulated camera-arm-gripper robot goes through two
phases:
(1) "Intrinsic phase": in a first long (15ML simulation steps) learning
phase the robot should acquire sensorimotor competence in a fully
autonomous way (no extrinsic reward functions, pre-wired knowledge on
objects and actions, etc.), on the basis of mechanisms such as free
exploration, curiosity, autonomous curriculum learning, intrinsic
motivations, and self-generated goals;
(2) "Extrinsic phase": the robot is tested with 50 "extrinsic goals",
unknown to the robot during the intrinsic phase, used to measure the
quality of the acquired knowledge that was autonomously acquired in the
first phase.

The key features of the competition as a benchmark for open-ended learning
are that:
(a) no information on tasks or on the specific domain can be given to the
robot during the first phase (full autonomy);
(b) the extrinsic phase allows a rigorous measure of the quality of the
knowledge that the robot autonomously acquired in the intrinsic phase on
the basis of the extrinsic goals representing a sample of all possible
goals/tasks that might be randomly drawn in the given environment; this
unique measure of the autonomously acquired knowledge facilitates the
comparison and improvement of models as it abstracts over the possible
"autonomy tricks" that different competing robots might use during the
intrinsic phase.

The benchmark is very challenging because during the intrinsic phase
it requires the robot to be able to do all these things at the same time in
a fully autonomous way:
(a) learn what objects are (e.g., location and identity) from raw pixel
images;
(b) learn motor skills, in particular, to get in contact with the objects
and move them (without moving the objects, it is difficult to distinguish
them from the rest of the environment);
(c) since during the intrinsic phase the environment is never reset (unless
objects fall off the table) the robot has to face continuously changing
environmental conditions.

The competition is based on a fully open-source software kit that relies on
a very fast 3D simulator (PyBullet) of the Kuka robotic arm with a gripper.
Moreover, it includes a fully functioning and modifiable "baseline” robot
architecture, based on modular well-commented software, to facilitate the
initial development of own models given the multiple challenges that have
to be faced at the same time. The kit thus allows a handy development of
your models on your computer before submitting them to the competition
website, and it can be used to make research on robot open-ended learning.

The top 3 winning teams will receive prizes, as indicated on the
competition website.

Important dates of the competition:
- 23/08/2021: competition started
- 18/01/2022: hands-on presentation of the competition
- 04/04/2022: hands-on micro-workshop at the Intrinsically Motivated
Open-ended Learning Workshop (IMOL 2022)
- 24/06/2022: competition ends
- 12-15/09/2022: presentation of winners at the International Conference on
Development and Learning (ICDL 2022; also launching REAL 2022)

For further details, please refer to the competition website:

https://eval.ai/web/challenges/challenge-page/1134/overview

.|.CS...|.......|...............|..|......US.|||.|||||.||.||||..|...|.......
Gianluca Baldassarre, Ph.D., Director of Research,
Laboratory of Embodied Natural and Artificial Intelligence,
Istituto di Scienze e Tecnologie della Cognizione,
Consiglio Nazionale delle Ricerche (LENAI-ISTC-CNR),
Via San Martino della Battaglia 44, I-00185 Roma, Italy.
Coordinator of LENAI Research Group: https://www.istc.cnr.it/it/group/locen
President of "Advanced School in AI": www.as-ai.org
President of "Associazione culturale science2mind": www.science2mind.org
Co-founder and R&D Officer of Spin-off CNR -
"Startup innovativa AI2Life s.r.l.": https://ai2life.com
E-mail: gianluca.baldassarre at istc.cnr.it
Web: http://www.istc.cnr.it/people/gianluca-baldassarre
Tel:  +39 06 44 595 231
Skype: gianluca.baldassarre
View of life: 'Learn from the past, live in(tensely) the present, dream for
the future'
Ultimate life mission: 'Serve humanity through core knowledge'
...CS.|||.||.|||.||..|.......|........|...US.|.|....||..|..|......|.........
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mailman.srv.cs.cmu.edu/pipermail/connectionists/attachments/20220113/7abb87c3/attachment.html>