Development of Pedestrian Behavior Model Taking Account of Intention (IROS2012)

Development  of  Pedestrian  Behavior  Model   Taking  Account  of  Intention Yusuke  Tamura1,  Phuoc  Dai  Le2,  Kentarou  Hitomi3,  Naiwala  P.  Chandrasiri3,   Takashi  Bando4,  Atsushi  Yamashita2,  Hajime  Asama2 1  Chuo  University 2  The  University  of  Tokyo 3  Toyota  InfoTechnology  Center,  Co.,  Ltd. 4  DENSO  Corp. 2012  IEEE/RSJ  International  Conference  on  Intelligent  Robots  and  Systems,  Vilamoura,  Portugal,  Oct.  8,  2012.

Motivation In  human-­‐robot  coexisting  environment... SAFETY  is  important. 2

Motivation In  human-­‐robot  coexisting  environment... SAFETY  is  important. Human  is  NOT  a  “moving  obstacle.” 2

Motivation In  human-­‐robot  coexisting  environment... SAFETY  is  important. Human  is  NOT  a  “moving  obstacle.” PREDICTION  of  human  behaviors  is  necessary. 2

Related  works  and  objective Pedestrian  model... -­‐  Social  force  model  [Helbing  1995][Helbing  2000] 3

Related  works  and  objective Pedestrian  model... -­‐  Social  force  model  [Helbing  1995][Helbing  2000] Unnatural  behavior 3

Related  works  and  objective Pedestrian  model... -­‐  Social  force  model  [Helbing  1995][Helbing  2000] Unnatural  behavior Objective Pedestrian  model:  producing  humanlike  behaviors   by  considering  pedestrian’s  intention 3

Basic  concept Create  an  appropriate  subgoal  according  to  the  pedestrian’s   attempted  behavior. Behavior 4

Basic  concept Create  an  appropriate  subgoal  according  to  the  pedestrian’s   attempted  behavior. Behavior Subgoal    creation 4

Basic  concept Create  an  appropriate  subgoal  according  to  the  pedestrian’s   attempted  behavior. Behavior Subgoal    creation 4

Basic  concept Create  an  appropriate  subgoal  according  to  the  pedestrian’s   attempted  behavior. Behavior Subgoal    creation 4

Basic  concept Create  an  appropriate  subgoal  according  to  the  pedestrian’s   attempted  behavior. Behavior Subgoal    creation 4

Proposed  pedestrian  model 5

Proposed  pedestrian  model 5

Proposed  pedestrian  model 5

Proposed  pedestrian  model 5

Proposed  pedestrian  model 5

Intention  transition  &  Behavior  selection Intention Behavior Free  Walk Avoid Follow 6

Intention  transition  &  Behavior  selection Intention Behavior Free  Walk Avoid Follow 6

Intention  transition  &  Behavior  selection Intention Behavior Free  Walk Avoid Follow 6

Intention  transition  &  Behavior  selection Intention Behavior Free  Walk Avoid Follow 6

Intention:  Free  walk  |  Behavior:  Free  walk Behavior  is  based  on  the  social  force  model  [Helbing  1995]. Repulsive  force  from  obstacles Repulsive  force  from   Acceleration  force  towards  Goal 7

Warning  area Pedestrian 8

Warning  area Pedestrian field  of  view 8

Warning  area Pedestrian desired  direction with  ranges field  of  view 8

Warning  area warning  area Pedestrian desired  direction with  ranges field  of  view 8

Intention  transition:  Free  walk  }}} warning  area Free  walk 9

Intention  transition:  Free  walk  }}} Free  walk 9

Intention  transition:  Free  walk  }}} Free  walk Avoid  or  Follow 9

Intention  transition:  Free  walk  }}} Free  walk Avoid Avoid  or  Follow Follow 9

Intention  transition:  Free  walk  }}} Free  walk Avoid Avoid  or  Follow Follow 9

Intention  transition:  Free  walk  }}} Free  walk Avoid Avoid  or  Follow Follow 9

Intention  transition:  Free  walk  }}} Free  walk Avoid Avoid  or  Follow Follow 9

Intention:  Follow  |  Behavior:  Follow Repulsive  force  from  obstacles Repulsive  force  from              (                                                    ) Acceleration  force  towards  Subgoal 10

Frustration Frustration Following  somebody  intensifies  the  pedestrian’s  frustration. Intention:  Follow   11

Frustration Frustration Following  somebody  intensifies  the  pedestrian’s  frustration. Intention:  Follow   11

Frustration Frustration Following  somebody  intensifies  the  pedestrian’s  frustration. Intention:  Follow   }}}  Avoid 11

Frustration Frustration Following  somebody  intensifies  the  pedestrian’s  frustration. Intention:  Follow   }}}  Avoid 11

Intention:  Avoid  |  Behavior:  Avoid   Repulsive  force  from  obstacles Repulsive  force  from             Acceleration  force  towards  Subgoal                    or   12

Prediction  for  avoidance 13

Prediction  for  avoidance 13

Prediction  for  avoidance 13

Prediction  for  avoidance 13

Prediction  for  avoidance 13

Prediction  for  avoidance BAD  subgoal 13

Prediction  for  avoidance BAD  subgoal 13

Prediction  for  avoidance BAD  subgoal 13

Prediction  for  avoidance BAD  subgoal GOOD  subgoal SELECTED 13

Prediction  for  avoidance BAD  subgoal GOOD  subgoal SELECTED If  both  candidates  are  good,            will  select  the  nearer  subgoal. 13

Prediction  for  avoidance BAD  subgoal GOOD  subgoal SELECTED If  both  candidates  are  good,            will  select  the  nearer  subgoal. If  both  candidates  are  bad,            will  be  forced  to  follow. 13

Experiments Comparison  between  simulated  and  observed  pedestrian  behaviors. Participants:  6  healthy  volunteers (5  men,  1  woman;  23-­‐29yrs.)            :  participant  (walk  to  G)            :  cooperator  (walk  to  R  or  C  or  L) (1) Scenario: (2) 1:          walks  to  G  while  passing            who        walks  straight  in  opposite  direction. 2:          walks  in  a  hurried  pace  to  G,  and                  walks  with  normal  speed  to  R  or   (3)        C  or  L. 3:          walks  in  a  hurried  pace  to  G,  and                walks  with  normal  speed  to  R  or          C  or  L. L R L R 14

Observation  and  Simulation Observation Measurement:   2  LRFs  (UTM-­‐30LX;  Hokuyo  Automatic,  Co.,  Ltd.) Height:  0.87m Frame  rate:  10fps Num.  of  trials: 6  participant  x  9  trials  (3  times  x  3  goals  of          )  for  each  scenario Simulation Environment: Input: Mac  OSX  10.6.8 C++  /  OpenGL Environmental  information,  observed  cooperator          ’s  trajectory Output: Trajectory  of  the  simulated  agent 15

Comparison  between  simulation  and  observation (e.g.)  Scenario  3  (        walked  to  C)   16

Comparison  between  simulation  and  observation (e.g.)  Scenario  3  (        walked  to  C)   16

Results Comparison  between  the  proposed  model  and  social  force  model Trajectory  Error  (m) 2.0 1.5 ** ** ** 1.0 0.5 ** p<0.01 0 (a) Scenario  1 (b) Scenario  2 (c) Scenario  3 paired  t-­‐test 17

Conclusion Pedestrian  behavior  model  considering  pedestrian’s  intention. -­‐  Subgoal  creation  based  on  the  intention  transition -­‐  Behavior  selection  based  on  the  prediction Proposed  pedestrian  model  can  produce  more  humanlike   behaviors  than  the  social  force  model. Future  works -­‐  A  Large  number  of  pedestrians -­‐  Online  parameter  estimation 18

20

Free  walk  }}} If  no  other  pedestrian  in  the  warning  area                  free  walk If  other  pedestrian  β  penetrates  the  warning  area  ... If  the  traveling  directions  of  α  and  β  are  roughly  opposite  ... avoid If  their  speed  difference  is  low  or  α  moves  slower  than  β  ... follow If  α  moves  much  faster  than  β  ... avoid 21

Intention:  Avoid  |  Behavior:  Follow Repulsive  force  from  obstacles Repulsive  force  from Acceleration  force  towards  Subgoal 22

Parameter  identification 8  parameters  were  determined  based  on  observation  experiments. Participants:  6  healthy  volunteers (5  men,  1  woman;  23-­‐29yrs.) Measurement:  2  LRFs UTM-­‐30LX  (Hokuyo  Automatic,  Co.,  Ltd.) Height:  0.87m Frame  rate:  10fps Obstacle  &  Robot:   ZEN  (Ritecs,  Inc.) Size:  0.45  x  0.45  x  0.75  m (a)  Desired  speed 6  times  for  each  condition. (b)  Warning  area  /  Subgoal  for  avoidance 6  times  for  each  obstacle  position v=0.3 or 0.5 or 0.8 m/s (c)  Subgoal  for  follow 3  times  for  each  combination  of  the  robot’s start  position  and  speed 23

Parameter  identification 8  parameters  were  determined  based  on  observation  experiments. A B C D E F                                    (m/s) 1.4 1.3 1.2 1.5 1.5 1.2                                    (m/s) 2.2 1.9 2.0 2.0 2.3 1.7                                        (m) 8.9 8.8 8.9 9.0 9.0 9.1                                        (m) 1.04 1.26 0.73 0.86 0.58 0.84                                        (m) 0.26 -­‐0.98 0.37 -­‐0.08 1.60 -­‐0.34                                        (m) 2.1 2.5 1.5 1.7 1.2 1.7                                        (m) 2.1 2.3 2.0 2.4 1.5 2.1                                        (m) 1.0 1.2 1.0 1.2 0.76 1.0 24