Gemini Robotics 2 brings whole body intelligence to robots — Transcript

Full transcript

1. 0:00How are you today, Apollo?
2. 0:01>> I'm a little nervous. Is that a
3. 0:02microphone?
4. 0:10>> [music]
5. 0:15[music]
6. 0:16>> For the robot, dealing with challenges
7. 0:18of the actual physical world is
8. 0:20borderline impossible.
9. 0:23>> [music]
10. 0:23>> But for us humans, it comes so naturally
11. 0:26and so intuitively that we almost don't
12. 0:28even think of it as intelligence.
13. 0:33>> You probably have seen all these amazing
14. 0:35videos of specialized humanoid robots
15. 0:37running, [music] jumping, backflipping.
16. 0:39The key difference in Gemini Robotics 2
17. 0:42>> [music]
18. 0:42>> is we aim to build a generalist robotics
19. 0:45model that's going to add a lot more
20. 0:46value if one robot can do a lot of
21. 0:49different tasks.
22. 0:50>> AI models offer [music] a way to
23. 0:53navigate the messy complexity of our
24. 0:57human environment.
25. 0:58>> The way to think about it is the brain
26. 1:00Gemini Robotics is what controls the
27. 1:03whole body of [music] the humanoid, the
28. 1:05delicate movements of the sharp hand and
29. 1:07the grippers of the dual.
30. 1:09>> We focus on three main things for this
31. 1:11release. [music] The first one is whole
32. 1:13body control. We are very used to move
33. 1:16our body in a very coordinated [music]
34. 1:17manner, but it's not true for robots.
35. 1:20Even though this feels a very simple
36. 1:22task for us, but for robots, I need to
37. 1:24make a many, many decisions across
38. 1:25[music] the entire body to achieve this.
39. 1:28The second thing is about dexterous
40. 1:30hands and particular [music]
41. 1:32manipulations of objects way beyond just
42. 1:34pick and place.
43. 1:36>> As a field, we're humbled
44. 1:38>> [music]
45. 1:38>> by things like screwing a light bulb
46. 1:40where you're interacting with smaller
47. 1:43objects, and those are really hard
48. 1:45problems [music]
49. 1:46that we are trying to solve.
50. 1:49>> When we first [music] came up with the
51. 1:51trash bag task in particular, several
52. 1:53people did think it was impossible. You
53. 1:55don't think about how to drive separate
54. 1:57[music] joints when you operate your
55. 1:59hand, but that's what we're asking these
56. 2:01AI models to do.
57. 2:03>> And for a thing [music] is being Gemini
58. 2:04Robotics too, we add a feature that
59. 2:07enables multiple robots to collaborate
60. 2:09[music] to accomplish the same task
61. 2:11simultaneously.
62. 2:12>> Hey duo, get all tools in the bin, close
63. 2:15the kit, and put the [music] kit back
64. 2:17into the bin.
65. 2:18>> On it.
66. 2:19>> Each of these two robots is running that
67. 2:21same stack, [music] and so rather than
68. 2:22having one robot network that controls
69. 2:24both robots, they have their own copy,
70. 2:25and they're [music] each doing their own
71. 2:26individual thinking, and they're
72. 2:28actually orchestrating through
73. 2:29reasoning.
74. 2:30>> Awesome precision, duo. Keep going.
75. 2:33>> We're really after intelligence and
76. 2:35generality. Being able to react to the
77. 2:37scene changing, [music] that is where
78. 2:39the intelligence comes in.
79. 2:42>> I'm super excited about robotics. I
80. 2:44think uh it's moving really fast, and
81. 2:47[music] AI is really the missing piece
82. 2:49of the entire robotics puzzle.
83. 2:54>> It is quite exciting to be in the
84. 2:55spotlight.