We are very proud that Nicholas Guttenberg, Senior Research Scientist at GoodAI, recently won the Evocraft 2021 competition – the Minecraft Open-Endedness Challenge!
The competition, which was a sidetrack of the GECCO 2021 conference, focused on open-endedness in order to highlight the progress in algorithms that can create novel and increasingly complex artefacts.
Most experiments in open-endedness so far have focused on simple toy domains however, the organizers believe Minecraft was the “perfect environment,” as a testing ground due to its unlimited possibilities.
In Nicholas’ entry, he tried to connect the idea of open-ended discovery to the open-ended development of new composable skills, something thought to be central to the ability of Badger-like architectures to exhibit continual lifelong learning. As a result, he made a process that continually discoversnew redstone circuits with different input-output functionalities to contrast an evolutionary novelty search-based approach with one based on purposeful compositional design. You can find out more about his entry in the video below and find his code repository here.
The entries were judged on 5 criteria:
- Divergence
- Diversity
- Complexity
- Ecological interactions
- Life-Like properties
A more detailed description of the criteria can be found here and the entries were judged by a prestigious jury made up of:
- David Ha, Google Brain
- Lisa Soros, Cross Labs
- Lana Sinapayen, Sony Computer Science Labs
- Kenneth Stanley, OpenAI
- Joel Lehman, OpenAI
As the winner, Nicholas won $1000, runners-up in the competition were two teams:
- Mayalen Etcheverry, Bert Chan, Clément Moulin-Frier, and Pierre-Yves Oudeyer. You can see their video here and their code repository here.
- Hugo Cisneros, Josef Sivic, and Tomas Mikolov. You can see their video here and their code repository here.
You can find out more details about the competition as well as the conference here.
Leave a comment