Context-aware Decision Making for Maze Solving

Transcription

1 RiTA 2012, Gwangju, Korea Context-aware Decision Making for Maze Solving Robot Inetelligence Technology Lab, KAIST Sheir Afgen Zaheer and Jong-Hwan Kim {sheir,

2 Contents 1. Introduction 2. Application setup 3. Problem Formulation 4. Context-aware decision making framework 5. Behavior Selection 6. Results 7. Conclusion 2

3 Introduction Different algorithms have been proposed for solving mazes with a mobile robot However, most of them require a pre-run to map the environment This paper proposed a context-aware decision making framework that enables the robot to solve an unknown maze. Only known things were the size and the position of the destination We compared our results with the wall following algorithm for maze solving 3

4 Application Setup The application setup consisted of a Khepera robot in a maze. The robot had Proximity sensors for detecting walls of the maze (Sensory Range 50 cm) Encoders for odometric and heading information Simulation Platform KiKS (T. Storm, 2010) 4

5 Problem Formulation Behavior selection was performed by evaluation of current context. Local situation position of walls surrounding the robot Position of robot in the maze For each context there was a small lists of behavior to choose from. The direction that robot should go to. Available behaviors were subjected to two criteria evaluation. The evaluation criteria for available behaviors were: The proximity of next obstacle in the direction of executed behavior (ΔD) The effect of behavior on the resulting distance from the destination point (d o ) 5

6 Decision Making Framework 6

7 Behavior Selection The degrees of consideration (Preferences) for each criterion were represented by λ-fuzzy measures. Since both criteria were uncorrelated, probability measures (λ =0) were used. g(δd)=0.8 (D C /D T *0.6) g(d o )=1 - g(δd) Fuzzy integral was finally used for global evaluation of each candidate behavior. (I. Gilboa, 1994) where n is the number of criteria, h(.) is the partial evaluation value, and Ei is the subset of the criteria set X consisting xi and all others that have a higher partial evaluation value than xi. 7

8 Behavior Selection Partial Evaluations for each criterion were defined in terms as: where PVc is the current proximity value measured by the ultrasonic sensor, and PVmax is the maximum proximity value when the sensor is touching the obstacle. 8

9 Results 60 x 60 cm maze 40 x 40 cm maze 9

10 Results 100 x 100 cm maze 10

11 Results Wall Following Algorithm Comparison with Wall following algorithm in terms of average solving time 11

12 Conclusions This paper proposed a context-aware decision making framework for a mobile robot to find its way through the maze. Simulations with various maze sizes were conducted. The proposed method for maze solving leads to a quick and successful solution but in its current form, it cannot be claimed as the optimal method. In future, incorporating a learning module can help the robot to regain its track if it gets lost when dealing with bigger, more complex non-linear mazes 12