ROILA : RObot Interaction LAnguage

The number of robots in our society is increasing rapidly. The number of service robots that interact with everyday people already outnumbers industrial robots. The easiest way to communicate with these service robots, such as Roomba or Nao, would be natural speech. However, the limitations prevailing in current speech recognition technology for natural language is a major obstacle behind the unanimous acceptance of speech interaction for robots. Current speech recognition technology is at times not good enough for it to be deployed in natural environments, where the ambience influences its performance. Moreover, state-of-art automatic speech recognition has not advanced far enough for most applications, partly due to the inherent properties of natural languages that make them difficult for a machine to recognize. Examples are ambiguity in context and homophones (words that sound the same but have different meanings). As a consequence of the prior discussed problems at times miscommunication occurs between the user and robot. The mismatch between humans’ expectations and the abilities of interactive robots often results in frustration for the user. Palm Inc. faced a similar problem with hand writing recognition for their handheld computers. They invented Graffiti, an artificial alphabet, that was easy to learn and easy for the computer to recognize. Our Robot Interaction Language (ROILA) takes a similar approach by offering a speech recognition friendly artificial language that is easy to learn for humans and easy to understand for robots with an ultimate goal of outperforming natural language in terms of speech recognition accuracy. There exist numerous artificial languages, Esperanto for example; but to the best of our knowledge these artificial languages were not designed to optimize human machine/robot interaction but rather to improve human-human communication. The design of ROILA was an iterative process having iterations within each step. It started off with a linguistic overview of a pre-selection of existing artificial languages across the dimensions of morphology (grammar) and phonology (the sounds of the language). The artificial languages were also analyzed in comparison to natural languages. The overview resulted in a number of linguistic trends that we would carefully incorporate in the design of ROILA with the claim that whatever linguistic features are common amongst these exist- ing languages would be easier to learn if they are made part of ROILA. The actual construction of the ROILA language began with the composition of its vocabulary. A genetic algorithm was implemented which generated the best fit vocabulary. In principle, the words of this vocabulary would have the least likelihood of being confused with each other and therefore be easy to recognize for the speech recognizer. Experimental evaluations were conducted on the vocabulary to determine its recognition accuracy. The results of these experiments were used to refine the vocabulary. The third phase of the design was the design of the grammar. Using the questions, options, and criteria (QOC) technique, rational decisions were made regarding the selection of grammatical markings. Recognition accuracy and ease of human learnability were two important criteria. In the end we drafted a simple grammar that did not have irregularities or exceptions in its rules and markings were represented by adding isolated words rather than inflecting existing words of a sentence. As a conclusion to the design phase and also as a proof of concept we designed an initial prototype of ROILA by using the LEGO Mindstorms NXT platform. ROILA was demonstrated in use to instruct a LEGO robot to navigate in its environment, analogous to the principles of the turtle robot. As a final evaluation of ROILA we conducted a large scale experiment of the language. ROILA was exposed to Dutch high school students who spent three weeks learning and practicing the language. A ROILA curriculum was carefully designed for the students to aid them in their learning both in school and at home. In-school learning was more interactive and hands on as the students tested their ROILA skills by speaking to and playing with LEGO robots. At the end of the curriculum the students attempted a ROILA proficiency test and if successful they were invited to play a complete game with a LEGO robot. Throughout the whole learning process, subjective and objective experiences of the students was measured to determine if indeed ROILA was easy to learn for the students and easy to recognize for the machine. Our results indicate that ROILA was deemed to have a better recognition accuracy than English and that it was preferred more by the students in comparison to English as their language of choice while interacting with LEGO Mindstorms robots.