1
00:00:00,691 --> 00:00:01,969
I was only nine

2
00:00:01,993 --> 00:00:06,916
when my grandfather first described to me
the horrors he witnessed six years earlier

3
00:00:06,940 --> 00:00:09,844
when human stampedes killed 39 people

4
00:00:09,868 --> 00:00:11,884
in our hometown of Nashik, India.

5
00:00:12,461 --> 00:00:15,750
It was during the 2003 Nashik Kumbh Mela,

6
00:00:15,774 --> 00:00:18,746
one of the world's largest
religious gatherings.

7
00:00:19,213 --> 00:00:22,943
Every 12 years,
over 30 million Hindu worshippers

8
00:00:22,967 --> 00:00:24,546
descend upon our city --

9
00:00:24,570 --> 00:00:27,410
which is built
only for 1.5 million people --

10
00:00:27,434 --> 00:00:29,077
and stay for 45 days.

11
00:00:29,601 --> 00:00:32,672
The main purpose
is to wash away all their sins

12
00:00:32,696 --> 00:00:35,021
by bathing in the river Godavari.

13
00:00:35,807 --> 00:00:38,045
And stampedes may easily happen

14
00:00:38,069 --> 00:00:41,346
because a high-density crowd
moves at a slow pace.

15
00:00:42,002 --> 00:00:46,311
Apart from Nashik, this event happens
in three other places in India,

16
00:00:46,335 --> 00:00:48,054
with varying frequency,

17
00:00:48,078 --> 00:00:50,847
and between 2001 and 2014,

18
00:00:50,871 --> 00:00:56,369
over 2,400 lives have been lost
in stampedes at these events.

19
00:00:57,362 --> 00:00:59,219
What saddened me the most

20
00:00:59,243 --> 00:01:03,758
is seeing people around me
resigning to the city's fate

21
00:01:03,782 --> 00:01:07,607
in witnessing the seemingly
inevitable deaths of dozens

22
00:01:07,631 --> 00:01:08,908
at every Kumbh Mela.

23
00:01:09,448 --> 00:01:10,878
I sought to change this,

24
00:01:10,902 --> 00:01:13,877
and I thought, why can't I try
to find a solution to this?

25
00:01:13,901 --> 00:01:15,389
Because I knew it is wrong.

26
00:01:15,701 --> 00:01:19,153
Having learned coding at an early age
and being a maker,

27
00:01:19,177 --> 00:01:20,845
I considered the wild idea --

28
00:01:20,869 --> 00:01:22,057
(Laughter)

29
00:01:22,081 --> 00:01:23,466
[Makers always find a way]

30
00:01:23,490 --> 00:01:26,555
I considered the wild idea
of building a system

31
00:01:26,579 --> 00:01:29,041
that would help regulate
the flow of people

32
00:01:29,065 --> 00:01:32,054
and use it in the next Kumbh Mela in 2015,

33
00:01:32,078 --> 00:01:35,683
to have fewer stampedes
and, hopefully, fewer deaths.

34
00:01:36,177 --> 00:01:38,648
It seemed like a mission impossible,

35
00:01:38,672 --> 00:01:40,440
a dream too big,

36
00:01:40,464 --> 00:01:42,645
especially for a 15-year-old,

37
00:01:42,669 --> 00:01:46,208
yet that dream came true in 2015,

38
00:01:46,232 --> 00:01:49,925
when not only did we succeed

39
00:01:49,949 --> 00:01:52,901
in reducing the stampedes
and their intensity,

40
00:01:52,925 --> 00:01:55,102
but we marked 2015

41
00:01:55,126 --> 00:01:59,133
as the first Nashik Kumbh Mela
to have zero stampedes.

42
00:01:59,596 --> 00:02:04,934
(Applause)

43
00:02:04,958 --> 00:02:07,444
It was the first time in recorded history

44
00:02:07,468 --> 00:02:10,214
that this event
passed without any casualties.

45
00:02:10,888 --> 00:02:12,038
How did we do it?

46
00:02:12,623 --> 00:02:15,654
It all started when I joined
an innovation workshop

47
00:02:15,678 --> 00:02:18,354
by MIT Media Lab in 2014

48
00:02:18,378 --> 00:02:20,347
called the Kumbhathon

49
00:02:20,371 --> 00:02:24,775
that aimed at solving challenges
faced at the grand scale of Kumbh Mela.

50
00:02:25,784 --> 00:02:30,292
Now, we figured out to solve
the stampede problem,

51
00:02:30,316 --> 00:02:32,480
we wanted to know only three things:

52
00:02:32,504 --> 00:02:35,243
the number of people, the location,

53
00:02:35,267 --> 00:02:38,726
and the rate of the flow
of people per minute.

54
00:02:39,241 --> 00:02:43,370
So we started to look for technologies
that would help us get these three things.

55
00:02:44,091 --> 00:02:48,345
Can we distribute radio-frequency tokens
to identify people?

56
00:02:48,369 --> 00:02:51,782
We figured out that it would
be too expensive and impractical

57
00:02:51,806 --> 00:02:53,742
to distribute 30 million tags.

58
00:02:54,545 --> 00:02:57,926
Can you use CCTV cameras
with image-processing techniques?

59
00:02:57,950 --> 00:02:59,997
Again, too expensive for that scale,

60
00:03:00,021 --> 00:03:02,899
along with the disadvantages
of being non-portable

61
00:03:02,923 --> 00:03:05,741
and being completely useless
in the case of rain,

62
00:03:05,765 --> 00:03:08,147
which is a common thing
to happen in Kumbh Mela.

63
00:03:09,120 --> 00:03:11,445
Can we use cell phone tower data?

64
00:03:11,806 --> 00:03:14,273
It sounds like the perfect solution,

65
00:03:14,297 --> 00:03:16,070
but the funny part is,

66
00:03:16,094 --> 00:03:18,839
most of the people
do not carry cell phones

67
00:03:18,863 --> 00:03:20,347
in events like Kumbh Mela.

68
00:03:20,887 --> 00:03:24,601
Also, the data wouldn't have been
granular enough for us.

69
00:03:24,625 --> 00:03:27,078
So we wanted something that was real-time,

70
00:03:27,102 --> 00:03:30,419
low-cost, sturdy and waterproof,

71
00:03:30,443 --> 00:03:32,854
and it was easy to get
the data for processing.

72
00:03:33,672 --> 00:03:36,028
So we built Ashioto,

73
00:03:36,052 --> 00:03:38,137
meaning "footstep" in Japanese,

74
00:03:38,161 --> 00:03:42,346
as it consists of a portable mat
which has pressure sensors

75
00:03:42,370 --> 00:03:44,861
which can count the number
of people walking on it,

76
00:03:44,885 --> 00:03:47,317
and sends the data over the internet

77
00:03:47,341 --> 00:03:49,888
to the advanced data analysis
software we created.

78
00:03:50,483 --> 00:03:53,771
The possible errors,
like overcounting or double-stepping,

79
00:03:53,795 --> 00:03:56,445
were overcome using design interventions.

80
00:03:56,803 --> 00:04:00,329
The optimum breadth of the mat
was determined to be 18 inches,

81
00:04:00,353 --> 00:04:02,592
after we tested many different versions

82
00:04:02,616 --> 00:04:05,203
and observed the average
stride length of a person.

83
00:04:06,111 --> 00:04:09,016
Otherwise, people might step
over the sensor.

84
00:04:09,040 --> 00:04:11,612
We started with a proof of concept
built in three days,

85
00:04:11,636 --> 00:04:14,510
made out of cardboard and aluminum foil.

86
00:04:14,534 --> 00:04:15,535
(Laughter)

87
00:04:15,559 --> 00:04:17,933
It worked, for real.

88
00:04:17,957 --> 00:04:20,647
We built another one
with aluminum composite panels

89
00:04:20,671 --> 00:04:22,309
and piezoelectric plates,

90
00:04:22,333 --> 00:04:25,904
which are plates that generate a small
pulse of electricity under pressure.

91
00:04:26,959 --> 00:04:30,228
We tested this at 30 different
pilots in public,

92
00:04:30,252 --> 00:04:33,974
in crowded restaurants,
in malls, in temples, etc.,

93
00:04:33,998 --> 00:04:35,632
to see how people reacted.

94
00:04:36,720 --> 00:04:38,743
And people let us run these pilots

95
00:04:38,767 --> 00:04:45,243
because they were excited to see localites
work on problems for the city.

96
00:04:45,680 --> 00:04:48,799
I was 15 and my team members
were in their early 20s.

97
00:04:49,736 --> 00:04:54,227
When the sensors were colored,

98
00:04:54,251 --> 00:04:57,092
people would get scared
and would ask us questions like,

99
00:04:57,116 --> 00:04:59,323
"Will I get electrocuted
if I step on this?"

100
00:04:59,347 --> 00:05:00,348
(Laughter)

101
00:05:00,372 --> 00:05:04,744
Or, if it was very obvious that it was
an electronic sensor on the ground,

102
00:05:04,768 --> 00:05:06,261
they would just jump over it.

103
00:05:06,285 --> 00:05:07,361
(Laughter)

104
00:05:07,385 --> 00:05:10,449
So we decided to design
a cover for the sensor

105
00:05:10,473 --> 00:05:13,530
so that people don't have to worry
what it is on the ground.

106
00:05:13,554 --> 00:05:15,530
So after some experimentation,

107
00:05:15,554 --> 00:05:18,364
we decided to use an industrial sensor,

108
00:05:18,388 --> 00:05:21,357
used as a safety trigger
in hazardous areas

109
00:05:21,381 --> 00:05:22,532
as the sensor,

110
00:05:22,556 --> 00:05:24,637
and a black neoprene rubber sheet

111
00:05:24,661 --> 00:05:25,903
as the cover.

112
00:05:26,374 --> 00:05:29,191
Now, another added benefit
of using black rubber

113
00:05:29,215 --> 00:05:32,842
was that dust naturally
accumulates over the surface,

114
00:05:32,866 --> 00:05:35,516
eventually camouflaging it
with the ground.

115
00:05:35,540 --> 00:05:40,652
We also had to make sure that the sensor
is no higher than 12 millimeters.

116
00:05:41,206 --> 00:05:43,435
Otherwise, people might trip over it,

117
00:05:43,459 --> 00:05:45,364
which in itself would cause stampedes.

118
00:05:45,792 --> 00:05:49,579
(Laughter)

119
00:05:49,603 --> 00:05:50,965
We don't want that.

120
00:05:50,989 --> 00:05:52,778
(Laughter)

121
00:05:52,802 --> 00:05:56,881
So we were able to design a sensor
which was only 10 millimeters thick.

122
00:05:57,469 --> 00:05:59,709
Now the data is sent
to the server in real time,

123
00:05:59,733 --> 00:06:01,381
and a heat map is plotted,

124
00:06:01,405 --> 00:06:04,748
taking into account
all the active devices on the ground.

125
00:06:04,772 --> 00:06:09,057
The authorities could be alerted
if the crowd movement slowed down

126
00:06:09,081 --> 00:06:12,216
or if the crowd density
moved beyond a desired threshold.

127
00:06:12,992 --> 00:06:17,984
We installed five of these mats
in the Nashik Kumbh Mela 2015,

128
00:06:18,008 --> 00:06:20,398
and counted over half a million people

129
00:06:20,422 --> 00:06:22,104
in 18 hours,

130
00:06:22,128 --> 00:06:26,505
ensuring that the data was available
in real time at various checkpoints,

131
00:06:26,529 --> 00:06:28,405
ensuring a safe flow of people.

132
00:06:28,429 --> 00:06:31,874
Now, this system, eventually,
with other innovations,

133
00:06:31,898 --> 00:06:35,050
is what helped prevent stampedes
altogether at that festival.

134
00:06:35,585 --> 00:06:38,458
The code used by Ashioto during Kumbh Mela

135
00:06:38,482 --> 00:06:41,894
will soon be made publicly available,
free to use for anyone.

136
00:06:41,918 --> 00:06:43,895
I would be glad if someone used this code

137
00:06:43,919 --> 00:06:46,131
to make many more gatherings safer.

138
00:06:46,878 --> 00:06:48,804
Having succeeded at Kumbh Mela

139
00:06:48,828 --> 00:06:52,740
has inspired me to help others
who may also suffer from stampedes.

140
00:06:52,764 --> 00:06:55,573
The design of the system
makes it adaptable

141
00:06:55,597 --> 00:06:57,591
to pretty much any event

142
00:06:57,615 --> 00:07:01,411
that involves an organized
gathering of people.

143
00:07:01,435 --> 00:07:05,745
And my new dream is to improve,
adapt and deploy the system

144
00:07:05,769 --> 00:07:11,238
all over the world to prevent loss of life
and ensure a safe flow of people,

145
00:07:11,262 --> 00:07:14,049
because every human soul is precious,

146
00:07:14,073 --> 00:07:16,451
whether at concerts or sporting events,

147
00:07:16,475 --> 00:07:18,362
the Maha Kumbh Mela in Allahabad,

148
00:07:18,386 --> 00:07:19,837
the Hajj in Mecca,

149
00:07:19,861 --> 00:07:21,758
the Shia procession to Karbala

150
00:07:21,782 --> 00:07:23,266
or at the Vatican City.

151
00:07:23,893 --> 00:07:25,770
So what do you all think, can we do it?

152
00:07:25,794 --> 00:07:26,945
(Audience) Yes!

153
00:07:26,969 --> 00:07:28,120
Thank you.

154
00:07:28,144 --> 00:07:29,145
(Cheers)

155
00:07:29,169 --> 00:07:32,439
(Applause)