|
Briefing
|
- Robot Language – OpenAI researchers discovered that bots can develop own language, assigning random characters to objects, actions and each other
- Virtual Environment Training – Placed bots in virtual world where they accomplished tasks such as looking at or moving to location, and helping others reach landmarks
- Rewards and Collaboration-Based – Implemented reward system that gives each bot sum of everyone's rewards, encouraging collaboration to maximize rewards they can receive
- Non-Verbal Communication – Bots also learned to guide each other, pointing and pushing in intended direction when language communication is unavailable
- English Translation – Plans to translate language into English by putting bots that developed own language in contact with English-speaking bots
|
|
Accelerator
|
|
|
Sector
|
Information Technology
|
|
Organization
|
Open Source Robotics Foundation
|
|
Source
|
-
Abbeel, P., et. al., "Learning to communicate",
-
Metz, C., "It begins: bots are learning to chat in their own language",
-
Mordatch, I. and Abbeel, P., "Emergence of grounded compositional language in multi-agent populations",
-
AcceleratingBiz analysis
|
|
Original Publication Date
|
March 16, 2017
|
