Alignment in vision-based syntactic language games for teams of robots using stochastic regular grammars and reinforcement learning: The fully autonomous case and the human supervised case
Introduction
Language, at both lexical and syntactic levels, is one of the fundamental cognitive skills necessary for the development of advanced and intelligent multi-robot systems as it allows communication and cooperation among the individuals of a robotic group. In previous work, using the so called language games concept [1] which is partly inspired in the ideas of Wittgenstein and de Saussure about the public and conventional dimensions of linguistic meaning, we have applied on-line reinforcement learning algorithms to the self-emergence of a common lexicon in robot teams [2]. In that work we modeled the lexicon or vocabulary of each robot as a look-up-table mapping the referential meanings (i.e. the objects or situations and states of the environment) into symbols. In this paper we extend our previous work on multi-robot lexical alignment through language games into multi-robot syntactic alignment also through dialogic language games and we study how the syntactic alignment can be developed in two different situations. First, in a group which is made up exclusively of similar artificial robots and second in a group including a robot that acts like a human being in the sense that this special robot is endowed with a grammar and a language similar to a human speaker. We will refer to this robot as human. Having two different configurations we can study if the artificial syntactic alignment can be influenced or mediated by the human intervention.
Section snippets
Multi-robot syntactic alignment
Although we acknowledge the fundamental relevance of lexical competence concerning language use and meaning [3] we do also believe that compositional, structural or simply syntactic competence is vital for an agent to efficiently describe reality in a symbolic, linguistic way, so that in this paper we approach the syntactic alignment of a robot team by means of dialogic language games applying also on-line reinforcement learning algorithms. The remaining part of this paper is organized as
Syntactic alignment among robots
In a first step, alignment language games must be applied so that a team with robots gets a common lexicon for the objects present in the environment as well as a common lexicon for the spatial relationships (right, left, front, behind). The acquisition of the objects lexicon is performed through a fully autonomous interactive process in which the robots are able to converge to common names for the different objects perceived as sensory discriminant variables, typical of Pattern Recognition
Syntactic alignment among robots and a human
Including a special robot acting like a human only implies minor changes in the system. First, we enumerate these changes. Following sections explain them with greater details.
- 1.
The human agent is endowed with a grammar with the right production rules. Obviously, it does not need to learn the probabilities.
- 2.
Reinforcement signal must be adapted in order to include the human’s perspective. This change modifies essentially the language games developing.
Experimental results
The experimental work has been focused on moderate robot team sizes in the range from five to sixty robots and we have also experimented with scenes involving a few specific objects and their spatial relationships. More specifically, the scenes of the training data set contain four specific objects (book, pencil, glasses and ball) and two relations (left and right). It is important to observe that the introduction of more objects and spatial relations does not imply any significant change or
Conclusions and future work
The use of stochastic regular grammars with learning capabilities has been proposed for the syntactic alignment of a robot team. The probabilities associated to each rewriting rule of the regular grammar provide the robots’ language with a plasticity that allows the robots to change their sentences aimed at converging to a common language with optimal communicative efficiency. The experimental scenario contemplated in this paper is based on digital images of scenes of the blocks world type.
Darío Maravall received his M.Sc. degree in Telecommunication Engineering from the Universidad Politécnica de Madrid in 1978 and his Ph.D. degree from the same university in 1980. From 1980 to 1988 he was an Associate Professor of Computer Science and Cybernetics Engineering at the School of Telecommunication Engineering, Universidad Politécnica de Madrid. In 1988 he was promoted to a Full University Professor of Computer Science and Artificial Intelligence in the Computer Science Faculty,
References (5)
- et al.
Coordination of communication in robot teams by reinforcement learning
Robot. Auton. Syst.
(2012)
Cited by (2)
A competence-performance based model to develop a syntactic language for artificial agents
2016, Information SciencesCitation Excerpt :This is also true to some extent for Fig. 7, past 15 agents. A similar behavior was also observed in [18] and it could be consequence of the larger number of interactions for large team sizes during the language games. In summary, both Figs. 7 and 8 confirm that CE rises as rounds go by and this shows that reinforcement learning works and complements well the previous evolutionary process.
Evolution of shared grammars for describing simulated spatial scenes with grammatical evolution
2018, Genetic Programming and Evolvable Machines
Darío Maravall received his M.Sc. degree in Telecommunication Engineering from the Universidad Politécnica de Madrid in 1978 and his Ph.D. degree from the same university in 1980. From 1980 to 1988 he was an Associate Professor of Computer Science and Cybernetics Engineering at the School of Telecommunication Engineering, Universidad Politécnica de Madrid. In 1988 he was promoted to a Full University Professor of Computer Science and Artificial Intelligence in the Computer Science Faculty, Universidad Politécnica de Madrid. From 2000 to 2004 he was the Director of the Department of Artificial Intelligence, Universidad Politécnica de Madrid. His current research interests include pattern recognition, computer vision and cognitive autonomous robots. He has published extensively on these topics and has directed more than 20 funded projects, including a five-year R&D project for the automated inspection of wooden pallets using computer vision techniques and robotic mechanisms, with several operating plants in a number of European countries and in the US. As a result of this multinational project he holds a patent issued by the European Patent Office at The Hague, The Netherlands. He has acted as Technical Consultant for numerous private firms and is currently a Senior Researcher in the Computational Cognitive Robotics Group of the CAR (Centro of Automática y Robótica), an official Research Center belonging to the Universidad Politécnica de Madrid and to the CSIC (the Spanish National Council of Scientific Research) where his group is working on the development of heterogeneous multirobots systems (mainly formed by ugvs and uavs).
Jack Mario Mingo received the B.Sc. degree in Informatics Engineering in 1993 and the M.Sc. degree in Informatics Engineering from the Universidad Politécnica de Madrid in 2002. Currently, he has completed his Ph.D. at the same university. From 1991 he has been working as Programmer, Analyst, Project Manager and Databases Specialist in several projects related to financial, logistics and telecom services. In parallel with these activities he is a Part-Time Professor at the Technical School, Universidad Autónoma de Madrid. His current research interests include grammatical evolution, evolutionary computation and autonomous robots.
Javier de Lope Asiaín received the M.Sc. degree in Computer Science from the Universidad Politécnica de Madrid in 1994 and the Ph.D. degree from the same university in 1998. Currently, he is an Associate Professor in the Department of Applied Intelligent Systems at the Universidad Politécnica de Madrid. His current research interest is centered on the study, design and construction of modular robots and multi-robot systems, and in the development of control systems based on soft-computing techniques. He is currently leading a three-year R&D project for developing industrial robotics mechanisms which follow the guidelines of multi-robot systems and reconfigurable robotics. In the past he also worked on projects related to the computer-aided automatic driving by means of external cameras and range sensors and the design and control of humanoid and flying robots.
- 1
Tel.: +34 913367442.