Research>Enabling Environment

Infant Behavior Simulator

>> Japanese

Overview
Although a tendency of infant accident is constant for long years, nobody can develope effective measures. Because tools for clarifying infant behaviors and the cause of infant accident don't exist. For solving the problem, development of the tool for simulating infant accident by recreating infant behavior on a computer is important. We thougt that infant behaviors depend on situation of environment and developmental stage strongly, we are developing infant behavior simulator based on these data. We look forward to infant behavior simulator being applied on evaluation of safety at home or clarification of cause of accident by integrating the simulator and accident database that we will create.
Behavior Modeling
We classify the determinant factors of infant behavior into internal factors and external factors. Internal factors are factors related to an infant that infant developmental behavior, physiology and so on. External factors are factors an environment surrounding an infant. We modeled developmental behavior and function of environmental interest and behavior induction because these factors relate infant behaviors strongly.
Factor related to infant behavior
Developmental Behavior Model
Developmental Behavior Model is modeling relation of age and possible behaviors (ex. A infant can walk when he/she is one year old.) and transition of behaviors (ex. sit -> stand is possible, sit -> run is impossible). We modeled using "Denver II". "Denver II" is a check sheet that pediatricians use for checking infant developmental stage.
発達に伴い、行動のネットワークが複雑になる。
Environmental Model
  • Function for inducing infant interest

  • We measure infant behavior and analyze the data in parallel with development of infant behavior simulator. We got a knowledge that relation of distance and infant interest from the data. We modeled this function using this knowledge.
  • Function for inducing infant behavior

  • We modeled this function using behaviors we got behaviors from behavior measurement and instance of accident.
Behavior Generation
We can get lists of a behavior and probability using following method. For example, grab a ball 0.2, pull himself up using a table 0.6, climb up a chair 0.4.

行動生成手法 1. Computation of probability that interest to a object
2. Computation of probability that induced behaviors by a object
3. Computation of probability that next behaviors based on state transition
4. Computation of probability that next behaviors based on developmental stage
5. Computation of probability that next behaviors
Verification of infant behavior simulator at the present phase
We verified the simulator comparing result of simulation and the data of behavior measurement. We calculated following items for comparing.
Inclusion rate: Rate of behavior nomination which simulator outputs inclues a behavior which acctual behavior takes. Contribution ratio: Ratio of impact of change of threshold on inclusion rate. Threshold is set to probability of behavior.

Result
検証結果
In the graph, horizontal axis indicates threshold which is set to probability of a infant next behavior. Left vertical axis indicates inclusion rate, right one indicates contribution ratio. Inclusion rate becomes higher as threshold becomes lower. Because lowering threshold means behavior nomination includes behaviors of low probability, i.e. behaviors that simulator predicts infant doesn't take them often. Contribution ration becomes lower as threshold becomes lower. It means that behaviors that simulator predicts infant take them often have a big impact on inclusion rate, behaviors that simulator predicts infant doesn't take them often have a small impact on inclusion rate. That is to say, actual infant takes behaviors that simulator predicts high probability with high probability, conversely, actual infant takes behaviors that simulator predicts low probability with low probability.

References

  • K. Kitamura, Y. Nishida, N. Matsumoto, Y. Motomura, T. Yamanaka, H. Mizoguchi, "Development of Infant Behavior Simulator: Modeling Grasping Achievement Behavior Based on Developmental Behavior Model and Environmental Interest Induction Model," Journal of Robotics and Mechatronics, Vol.17, No.6, 2005 (in press)
  • A. Simo, K. Kitamura, Y. Nishida, "Children Accidents' Simulation: Planning the Emergent Situations," Proc. of the 4th International Conference on Interaction Design and Children," June 2005
  • A. Simo, K. Kitamura, Y. Nishida, "Behavior based Children Accidents' Simulation and Visualization: Planning the Emergent Situations," in Proceedings of the Fourth IASTED International Conference on Computational Intelligence (CI 2005), pp. 164-169, July 2005
  • Y. Nishida, K. Kitamura, Y. Motomura, A. Simo, T. Yamanaka, "Infant Behavior Simulation: Computational Approach to Infant Safety," in Proceedings of the 4th IARP/IEEE-RAS/EURON Workshop on Technical Challenges for Dependable Robots in Human Environments, T16-01(1)-(12), June 2005
  • A. Simo, K. Kitamura, Y. Nishida, "3D Visualization of Children Accidents Related Behavior: An Integrated Approach," in Proceedings of 10th International Conference on Virtual Systems and Multimedia, pp. 416-425, November 2004
  • Y. Nishida, K. Kitamura, Y. Motomura, H. Mizoguchi, "Infant Behavior Simulation Based on an Environmental Model and a Developmental Behavior Model," in Proceedings of 2004 IEEE International Conference on Systems, Man and Cybernetics (SMC '04), pp. 1555-1560, October 2004