| 研究実績の概要 |
In this period, another novel rule-base reduction method was developed, which is based on clustering
techniques. This method examines whole fuzzy rules, and tries to merge several rules into one rule. This way, simplifying the rule-base, which is the knowledge representation itself in this case. Compared to the previous rule-base reduction strategies in the FRIQ-learning method, which could only remove certain rules and leaving the remaining rules untouched, this method is capable of removing a group of rules and also substitute them with one rule, hence effectively replacing them. This way the knowledge stored in the form of fuzzy rules can be directly and more easily read by human experts. A new method capable of inserting existing expert knowledge as a base for FRI-RL was developed. When injecting existing partial knowledge-bases, the new method was able to correctly extend it, and when a complete knowledge-base was supplied, it was able to confirm its correctness. Additionally, the foundations have been laid of a possible interface between the FRIQ-learning framework and a complex behaviour simulation application (Strange Situation Test simulation). While working on the knowledge extraction from behaviour models, the need for automatically observing and evaluating behaviours in Human-Robot Interaction has emerged. Fundamentals of such system have been laid, and a proof-of-concept implementation was created.
|
| 今後の研究の推進方策 |
The research is continued as planned, with additions:
- Test and evaluate the developed clustering-based fuzzy rule-base reduction methods.
- The tuning of the SST simulation behaviour model to be able to satisfy the requirements of the real-life SST test.
- The design and implementation of connecting the attachment behaviour model to a real-life physical robot.
- Use the developed automatical behaviour recognition system to observe and evaluate real-life Human-Robot Interactions. For this, the capabilities of appropriate sensors and indoor localization systems should be investigated and tested
|
