Developing Robots That Can Learn Like Humans

Recently, there have been remarkable advances in artificial intelligence (AI). It is said that AI is superior to human intelligence in supervised learning where vast amounts of labeled data is used to perform specific tasks. However, it is considered that it is difficult to realize human-like intelligence using only supervised learning because all supervised labels cannot be obtained for all the sensory information required by robots.

Here, Tomoaki Nakamura of UEC, Tokyo, and colleagues are conducting research on the realization of robots that can acquire knowledge in a manner similar to human beings.

To this end the researcher believes that it is important for robots to understand their environment by structuring their own sensory information in an unsupervised manner.

Recently, Nakamura proposed an algorithm that enables robots to learn concepts and language [1,2]. The robots obtain multimodal information from objects, and linguistic information by communicating with others. Using this information, the algorithm allows robots to form object concepts and learn languages. Moreover, concepts that are learned by robots using this algorithm are compared with the corresponding human concepts and the similarities between them are shown [3].

Nakamura has also proposed a method for robots to learn motions by observing human motion [4]. Additionally, this method also allows robots to learn the rules of interaction by observing human interaction.

"Through this research, I would like to develop robots like humans," says Nakamura. "Additionally, I believe that this research will lead to a better understanding human intelligence."

Reference

• [1] Joe Nishihara, Tomoaki Nakamura, Takayuki Nagai, "Online Algorithm for Robots to Learn Object Concepts and Language Model", IEEE Transactions on Cognitive and Developmental Systems, Vol. 9, No. 3, pp. 255-268, 2017
• [2] Tomoaki Nakamura, and Takayuki Nagai, "Ensemble-of-Concept Models for Unsupervised Formation of Multiple Categories", IEEE Transactions on Cognitive and Developmental Systems, DOI:10.1109/TCDS.2017.2745502, 2017
• [3] Miyuki Funada, Tomoaki Nakamura, Takayuki Nagai and Masahide Kaneko, "Analysis of the Effect of Infant-Directed Speech on Mutual Learning of Concepts and Language Based on MLDA and Unsupervised Word Segmentation", IROS2017: Workshop on Machine Learning Methods for High-Level Cognitive Capabilities in Robotics, Sep. 2017
• [4] Tomoaki Nakamura,Takayuki Nagai, Daichi Mochihashi, Ichiro Kobayashi, Hideki Asoh and Masahide Kaneko, "Segmenting Continuous Motions with Hidden Semi-markov Models and Gaussian Processes", Frontiers in Neurorobotics, DOI:10.3389/fnbot.2017.00067, 2017
• Affiliations: Graduate School of Informatics and Engineering, The University of Electro-Communications, Tokyo.
• Website: http://www.naka-lab.org/
• Researcher Video Profiles: Tomoaki Nakamura Assistant Professor, Graduate School of Informatics and Engineering.