The goal of this project was to show that complex behavior patterns can be learnt by a system based on reinforcement learning. The specific task was to make AIBO, the Sony robot dog, learn complex behavior patterns based on interactions between humans and AIBO. The reinforcement learning system is taught by remote control, used by the human and connected to AIBO. To remember the learnt behavior sequences, a short-term memory of prior actions is used by AIBO. This paper demonstrates that it is possible to learn behavior sequences and the relationship of cause and effect in complex environments. The paper also shows that the system works in a natural environment, based on the interaction between humans and AIBO, learning the rewards and the means to reach them in parallel. AIBO is also able to pick up new behaviors instantly by using a method we call 'Instant learning'. The paper presents the methods for implementing such a system.
|出版ステータス||Published - 2004|
|イベント||SICE Annual Conference 2004 - Sapporo, Japan|
継続期間: 2004 8月 4 → 2004 8月 6
|Conference||SICE Annual Conference 2004|
|Period||04/8/4 → 04/8/6|
ASJC Scopus subject areas
- コンピュータ サイエンスの応用