Please refer to Lab Website for latest research topics.

Desgning a Neuromoduratory Neural Netwroks for Hand Reaching Movements Control in Unknown Viscous Force Fields

Regardless of complex, unknown, and dynamically-changing environments, living creatures can recognize situated environments and behave adaptively by theirselves in real-time. However it is impossible to prepare optimal motion trajectories with respect to every possible situation in advance. The key concept for realizing suitable environmental cognition and motor adaptation is a context-based elicitation of constraints which are canalizing well-suited sensorimotor coordination. For this aim, in this study, we propose a polymorphic neural networks model called CTRNN+NM (CTRNN with neuromodulatory bias). The proposed model is applied to two dimensional arm-reaching movement control in various viscous curl force fields. The model parameters were optimized by GA. Simulation results reveal that the proposed model inherits high robustness even though it is situated in unexperienced environment, which has same curl but different size of viscous force, since it evolved "how to adapt" instead of "how to move."

  1. Toshiyuki Kondo, Koji Ito: "A Neuromoduratory Neural Networks Model for Environmental Cognition and Motor Adaptation", Proceedings of IEEE World Congress on Computational Intelligence (WCCI2006), Vancouver, Canada, pp.9865-9870 (2006)

Periodic motion control by modulating CPG parameters based on time-series recognition

This paper proposed a computational model of the periodic motion control of physical controlled objects inspired by biological brain-motor systems. The proposed motion control model consists of two processing layers named "CPG layer" and "Dynamical Memory layer. Likewise biological central pattern generators (CPG) in spinal cord, the CPG layer plays a role in generating torque patterns for realizing periodic motions. On the contrary, the Dynamical Memory layer is a time-series pattern discriminator implemented by recurrent neural networks (RNN). By associating the time-series observation of system state with the optimized CPG parameters, the RNN can predictively modulate generating torque patterns by recalling well-suited CPG parameters according to the observed situation.

  1. Toshiyuki Kondo, Takanori Somei, Koji Ito: "A predictive constraints selection model for periodic motion pattern generation", Proceedings of 2004 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS'04), TP1-K2, pp.975-980, Sendai, Japan, (2004)

A Reinforcement Learning with Evolutionary State Recruitment Strategy for Autonomous Mobile Robots Control

In recent robotics fields, much attention has been focused on utilizing reinforcement learning (RL) for designing robot controllers, since environments where the robots will be situated in should be unpredictable for human designers in advance.

However there exist some difficulties. One of them is well known as "curse of dimensionality problem". Thus, in order to adopt RL for complicated systems, not only "adaptability" but also "computational efficiencies" should be taken into account.

The paper proposes an adaptive state recruitment strategy for NGnet-based actor-critic RL. The strategy enables the learning system to rearrange/divide its state space gradually according to the task complexity and the progress of learning. Some simulation results and real robot implementations showed the validity of the method.

  1. Toshiyuki Kondo, Koji Ito: "A Study on Designing Robot Controllers by Using Reinforcement Learning with Evolutionary State Recruitment Strategy", Lecture Notes in Computer Science 3141 -Biologically Inspired Approaches to Advanced Information Technology: First International Workshop BioADIT 2004 Lausanne Switzerland January 29-30 2004 Revised Selected Papers, Springer -Verlag Berlin Heideberg, pp.244-257, (2004)
  2. Toshiyuki Kondo, Koji Ito: "A Reinforcement Learning with Evolutionary State Recruitment Strategy for Autonomous Mobile Robots Control", Journal of Robotics and Autonomous Systems, vol.46, no.2, pp.111-124 Elsevier, (2004)

An Incremental Reinforcement Learning using Constraint Rules Extraction Mechanism for Autonomous Mobile Robots

A number of skillful robots have been developed in the last decade, however most of them can only demonstrate pre-programmed motions according to external stimuli. In contrast, humans can learn new motions in spite of their high dimensional sensors/actuators DOF (degrees of freedom). In human motor learning process, it can be hypothesized that the learner should actively constrain the DOF by him/her-self using some "learning skills", here referred to as ``constraint rules.''

In this paper, we propose a learning method for the autonomous mobile robot operated in unknown environments. In the method, not only training of sensorimotor coordinations, but also extraction/reuse mechanism of constraint rules are implemented. Through the results of simulations and real experiments of a mobile robot navigation, the validity of the proposed method is clarified.

  1. Toshiyuki Kondo, Norihiko Itoh, Koji Ito: "An Incremental Learning using Schema Extraction Mechanism for Autonomous Mobile Robot", Proceedings of 2003 IEEE International Symposium on Computational Intelligence in Robotics and Automation, CD-ROM, pp.1126-1131, Kobe, Japan, (2003)

Continual Human-Agent Interaction approach for Entertainment Robots

In the field of Robotics and Artificial Intelligence, there have been gradually increasing the works dealing with Human-agent Interactions (HAI), e.g. humanoid robot applications, pet robot developments, interactive teaching systems, affective computing and so on. However, most of these studies just aimed at developing intelligent agents, or referred only to unavoidable nature of humans, that is, we could implicitly adapt to artificial systems. One of the diffculties in this HAI framefork is how to keep a continual interaction between a human and an artificial agent, since human could apt to lose interest in the agent as time goes on. Due to actualize continual interactions between agents and the users, it seems to be crucial that both systems possess intrinsic/epigenetic adaptation abilities and the agent should have an evaluation criteria for estimating internal conditions (e.g. curiosity) of users. Based on this consideration, in this study, we discuss key functions for realizing a continual HAI, which the agent should possess.

  1. Yoshihisa Wakamatsu, Toshiyuki Kondo, Koji Ito: "A Proposal of Communication Design for Continual Human-Agent Interaction Using Natural Utterances", Proceedings of SICE Annual Conference 2003, WPI-17-5, Fukui, Japan, (2003)

Evolving behavior in devoloping robot bodies controlled by quasi-Hebbian neural networks

LEGO(TM) Robot Project targets evolving a suitable body shape and its nervous system (i.e. artificail neural network-based controller) for a light seeking robot. Using a grammar-based gene encoding method, we can save computational resources (length of genotype), in other words, it enables to evolve much faster than ordinary coding methods.

In the current situation, I revised "dynamically-rearranging neural networks" method much more useful. For example, instead of each synapse, each neuron has its own NM interpretation table, and then each synapse originated from same neuron uses the interpretation table. In addition, effective area of each NM is restricted within some small range (this is genetic parameter). It is expected this enables the emerged network to have more complicated structure and flexibility, redundancy. In this study, we are concerned with the interaction between three specific adaptive systems: evolutionary change by species, ontogenic change by an individual as it matures; and learning by the individual as it acquires experience. We present experiments in which a population of individuals, each grown from a single cell according to its particular genome, into an adult form corresponding to a simple robot body and NNet which allows it to function in and learn about its environment.

  1. Submitted to 7th Joint Symposium on Neural Computation 7 Apr 00

A Dynamically-rearranging Neural Network Approach for Evolutionary Robotics

Recently, Evolutionary Robotics approach has been attracting a lot of concerns in the field of robotics and artificial life. In this approach, neural networks are widely used to construct controllers for autonomous mobile agents, since they intrinsically have generalization, noise-tolerant abilities and so on. However, the followings are still open questions; 1) gap between simulated and real environments, 2) evolutionary and learning phase are completely separated, and 3) conflict between stability and evolvability/adaptability. In this article, we try to overcome these problems by incorporating the concept of dynamic rearrangement function of biological neural networks with the use of neuromodulators.

  1. Toshiyuki Kondo, Evolutionary design and behavior analysis of neuromodulatory neural networks for mobile robots control, Applied Soft Computing, Elsevier, in press
  2. Toshiyuki Kondo, Akio Ishiguro, Seiji Tokura, Yoshiki Uchikawa, Peter Eggenberger: "Realization of Robust Controllers in Evolutionary Robotics: A Dynamically-Rearranging Neural Network Approach", Proceedings of the 1999 Congress on Evolutionary Computation (CEC'99), Vol.1, pp.366-373, Washington D.C., USA, (1999)

Emergent Adaptation model inspired by Biological Immune System

Conventional arti.cial intelligent (AI) have been criticized for its brittleness under hostile/dynamic changing environments. Therefore, recently much attention has been focused on the reactive planning systems such as behavior-based AI. However, in the behavior-based AI approaches, how to construct a mechanism that realizes adequate arbitration among competence modules is still an open question. In this paper, we propose a new decentralized consensus-making system inspired from the biological immune system. And we apply our proposed method to behavior arbitration of an autonomous mobile robot as a practical example. To verify the feasibility of our method, we carry out some experiments.In addition, we propose an adaptation mechanism, and try to construct a suitable immune network for adequate action selection.

  1. Toshiyuki Kondo, Akio Ishiguro, Yuji Watanabe, Yasuhiro Shirai, Yoshiki Uchikawa: "An Evolutionary Construction of Immune Network-Based Behavior Arbitration Mechanism for Autonomous Mobile Robot", Electrical Engineering in Japan, Vol.123, No.3, pp.1-10, (1998)

yesterday:0 today:1 total:3111