Reachy Mini: An Open Journey Built Together with Hugging Face, Pollen Robotics, & Seeed Studio — Transcript

Full transcript

1. 0:03[music]
2. 0:10I'm Richi Mini. Born from Pollen
3. 0:12Robotics, powered by hugging face models
4. 0:14and mass-produced by C. Bringing open
5. 0:17robotics [music] to everyone.
6. 0:18>> Can you say exactly the same thing but
7. 0:20in Chinese now?
8. 0:24[music]
9. 0:25Poland robotics
10. 0:32[music]
11. 0:36you
12. 0:37>> has been quite an journey to reach this
13. 0:40level.
14. 0:41>> Yes.
15. 0:42>> And how many times have been flying from
16. 0:44France to Shenzhen
17. 0:46>> this year?
18. 0:47>> It's the second times and [music]
19. 0:49>> second time.
20. 0:50>> Second times. Yeah. And for two weeks
21. 0:52each Chinese
22. 0:53>> it's very rare [music] to see some
23. 0:55French working so hard.
24. 0:56>> Yeah.
25. 0:57>> Yeah.
26. 0:58>> And you you [music] like uh work hours
27. 1:01exceeding a lot of our engineers is you
28. 1:03stay in office all the time.
29. 1:05>> Even they're not to disturb you
30. 1:07>> because you always coding and you're not
31. 1:10like CEO you're like a fullstack
32. 1:12engineer.
33. 1:13Uh yeah and it's clear we are doing a
34. 1:16lot of thing at Pol and uh everyone is
35. 1:20really uh 100% on on this [music]
36. 1:23project on your team
37. 1:24>> and also um the timeline is so short
38. 1:28that we really don't have time to to do
39. 1:31another way.
40. 1:32>> Yeah. So we are pushing [music] the
41. 1:34limits.
42. 1:35>> Yeah.
43. 1:35>> How many months did we spend? Three
44. 1:38months.
45. 1:38>> Yeah. Just just a few months. Uh we
46. 1:42launched the robot in uh in July and it
47. 1:45was just a prototype with a 3D printing
48. 1:48part.
49. 1:48>> Yeah.
50. 1:49>> Uh
51. 1:51we planned to to have about [music]
52. 1:54100 or 200 units to to to sell uh few
53. 1:58months later. So it was all made in a
54. 2:02fully printing uh [music] spirit.
55. 2:04>> Yeah. But we reach actually more than
56. 2:073,000
57. 2:09>> [music]
58. 2:09>> uh sale in a in a week in in a week. In
59. 2:13a week
60. 2:15uh so we had to find another way to
61. 2:17produce a robot and so just uh [music]
62. 2:20one week after the launch we were at SID
63. 2:24uh to start working on on a way [music]
64. 2:27to transform this robot into a mass
65. 2:30production uh product.
66. 2:32>> Mhm. [music] Uh so actually uh seat
67. 2:35started to work on on the mass
68. 2:38production unit only August [music] and
69. 2:41now we are early November and the robot
70. 2:45is ready to go into production. So we
71. 2:48are
72. 2:50really grateful
73. 2:53uh to having met you to
74. 2:57uh to to do this project [music] because
75. 2:59I think we we won't be able to to
76. 3:02achieve that level of uh quality and
77. 3:06speed uh to to [music] deliver this
78. 3:09product.
79. 3:10>> We are very excited as well to be part
80. 3:13of this project. So I think I see Rich
81. 3:16Mini quite a lot early this year. I
82. 3:21think we do [music] some renderings from
83. 3:22Poland Robotics, but I know Polar
84. 3:24Robotics for many years ago. We have the
85. 3:27bigger that Richie
86. 3:29>> and uh [music] I like the design because
87. 3:31I'm big fans of Gibli.
88. 3:33>> Yeah. And this uh kind of big eye, small
89. 3:36eye and all the all the emotional part
90. 3:39of this robot is really uh
91. 3:41differentiating it from all the other
92. 3:44[music] uh human noise and I think the
93. 3:46team has been totally geared up towards
94. 3:49this goal and uh we [music] in
95. 3:51internally we turn that into priority S
96. 3:54which means the highest standards
97. 3:56[music] and everybody has to like finish
98. 3:59any task relating to it at prop priority
99. 4:02and we We have been doing like uh robot
100. 4:05arms from maybe 12 [music] years ago to
101. 4:08assemble them we work with u arm but for
102. 4:11many years we [music] do small robots
103. 4:13this is one of the biggest one we have
104. 4:15been doing okay
105. 4:16>> it's [music] learning curve it's a bit
106. 4:18challenging to for us as well if we
107. 4:20break it down I think u a lot of things
108. 4:23we are familiar like for the speaker we
109. 4:25have done on the uh sound part [music]
110. 4:28for 10 years
111. 4:30>> and uh but combined with robot how do
112. 4:32you den noiseise that [music] how do you
113. 4:34resolve the u like when it's moving how
114. 4:37do you tell
115. 4:38>> uh how do you interrupt that with your
116. 4:39[music] voice is I think there's a lot
117. 4:41of
118. 4:42>> yeah new exciting challenges are our
119. 4:45engineers like that yeah and that's was
120. 4:48one of the big challenge uh the sound
121. 4:50the voice input of robot
122. 4:53>> actually voice is [music] the only way
123. 4:55to interact with the robot
124. 4:57>> okay
125. 4:57>> because there is no screen no no
126. 5:01keyboard and so on so [music] the only
127. 5:02way to have direct interaction with
128. 5:04robot is not is a voice
129. 5:07uh [music] and we didn't have this uh
130. 5:09speciality in our team. It's also why we
131. 5:13[music] started to work with with
132. 5:14speaker and trying to find a way to
133. 5:17[music] make it work with your
134. 5:19technology
135. 5:20>> and it's also that drive us toward a
136. 5:23collaboration with SID because you
137. 5:26really [music]
138. 5:27know way better than us how to achieve
139. 5:30uh to do microphone array and the noise
140. 5:33and at the end uh I think we are pretty
141. 5:37happy with the result because we the
142. 5:40robot can really hear wells the [music]
143. 5:43voice even when you're far far away and
144. 5:46that's really challenging because there
145. 5:48is a lot of motors inside the robot
146. 5:50>> especially antenna just behind
147. 5:53>> behind the microphone
148. 5:54>> the microphone
149. 5:55>> behind the microphone so that's really
150. 5:57not [music] a great environment to have
151. 5:59voice detection uh but at the end after
152. 6:03a few iteration
153. 6:05>> I think it's now really convincing
154. 6:08>> very very happy to know this So, and
155. 6:11when I see this robot, it's going all
156. 6:13the way. This is the first injection
157. 6:15modes. Yeah. And then the next time
158. 6:19people can like hank [music] or paint
159. 6:21them as different colors, different
160. 6:23skins or different decorations.
161. 6:25>> Yeah. It's also why we took it really
162. 6:28simple simple white shape. So you can
163. 6:31white on it, add stickers, uh or also as
164. 6:35a robot is open source, you can 3D print
165. 6:38parts to replace [music] them or to add
166. 6:40them. So, for example, on on Halloween a
167. 6:44few weeks ago, we did a full
168. 6:46>> the pumpkin one,
169. 6:47>> a pumpkin one, [music]
170. 6:48>> and it was a a full costume where you
171. 6:50can add the pumpkin on on a ship. And it
172. 6:54took us only uh 24 hours between the the
173. 6:58idea to do that and the actual uh
174. 7:01actually having the robot [music]
175. 7:03ready
176. 7:04>> uh and just 3D printing a big bump to
177. 7:08put around the robot.
178. 7:10I think [music] as long as we tackle the
179. 7:12manufacturing difficulties to have the
180. 7:15first thousands of batches shipped, I
181. 7:17think [music] the next step is uh to
182. 7:19introduce to the maker communities.
183. 7:22>> I think a lot of makers will be excited
184. 7:24including myself to assemble [music] one
185. 7:26and also to add more functions to that.
186. 7:29So super humanize personalize my Richi
187. 7:33and maybe in near future they can have
188. 7:35dramas can have plays beyond the like
189. 7:38companioning robots they should meet my
190. 7:40uh avata or physique [music]
191. 7:43uh how puppet agent.
192. 7:46>> Yeah I think really now the now the
193. 7:49hardware [music] is really good and able
194. 7:52to be mass produced. [music] Now the
195. 7:54challenge is to have a large community
196. 7:57of developers.
197. 7:58>> Yeah. and AI developers that can that
198. 8:02but is not really roboticist [music]
199. 8:04because the robot is really easy to use
200. 8:07if you do know nothing about robotics.
201. 8:10>> Mhm.
202. 8:10>> The app [music] the API is quite simple.
203. 8:13It's uh made with Python SDK and so on.
204. 8:16[music] It's really accessible. Um so
205. 8:19any software engineer or AI developer
206. 8:22can use this robot to [music] build
207. 8:24application and push application on
208. 8:28internet and especially [music] on ugly
209. 8:30face spaces and then we have on the
210. 8:33robot software to
211. 8:36um download those application easily
212. 8:39just with one click. So if a developer
213. 8:43is building an application, they can
214. 8:45push [music] it on face and then any
215. 8:48anyone who has a rich can download it
216. 8:51and try it and we are [music] really
217. 8:53excited about what can happen with this
218. 8:55kind of stuff because it's really easy
219. 8:57to push new app easy to download app and
220. 9:01so
221. 9:02I hope we will see a lot of creation
222. 9:06around this robot and people testing
223. 9:07[music]
224. 9:08sharing and so on and at the end we can
225. 9:12build really [music]
226. 9:13both playful and useful apps.
227. 9:16>> Sure. I think whenever this is out
228. 9:18[music] in the market you can easily get
229. 9:20them uh people with different background
230. 9:23can upgrade or change it very nicely. So
231. 9:28all the software developers [music] can
232. 9:30bring their new soul and all the makers
233. 9:33can add new like maybe sensors different
234. 9:35[music] functions with some like
235. 9:36activators. Uh how is it to have
236. 9:39different like extension kind of actual
237. 9:42like arm
238. 9:43>> ah
239. 9:44>> like the cat to welcome you?
240. 9:46>> We we let [music] some part to to do
241. 9:49that. So the the robot has a USB port.
242. 9:52Uh so you can plug [music] anything you
243. 9:54need USB.
244. 9:56>> Okay. So this is the host.
245. 9:58>> It can be it can be the host.
246. 9:59>> Yeah. Yeah. So basically we can connect
247. 10:01[music] the reach Mini to any like
248. 10:03Arduino or SH based projects.
249. 10:06>> To have extra arm or extra [music] like
250. 10:10uh uh gadgets.
251. 10:12>> Yes.
252. 10:13>> Yeah. And on the Raspberry IP version,
253. 10:16this one running there,
254. 10:18>> [music]
255. 10:18>> uh you have a battery, so the robot can
256. 10:21work for about 2 hours, [music] but also
257. 10:24he has a Bluetooth. So you can add a
258. 10:26Bluetooth modules if you want.
259. 10:28>> Wow. connect to the robot. So nice to
260. 10:32show way to [music] add new sensor on
261. 10:34the robot.
262. 10:35>> Yeah.
263. 10:35>> And I see there's quite some room inside
264. 10:37the the body and the head.
265. 10:40>> So if you can [music] more like a
266. 10:42sensors, cameras, even like uh brass
267. 10:45kind of MMO wave into [music] it. Can't
268. 10:48wait to see its evolutions.
269. 10:50>> Yeah. Now [music]
270. 10:52she can hear you, hear us, and talk with
271. 10:55us.
272. 10:56>> Hey there. Loud and clear. ready to chat
273. 10:59about how we're making things happen
274. 11:00with mass production.
275. 11:03>> Yeah. So, Richie,
276. 11:06can you uh show us how you can move?
277. 11:12>> Here we go. Swaying side to side with a
278. 11:15smooth move.
279. 11:16>> Yeah. So, I I really think what's great
280. 11:18with Rich Mini is that it's a some kind
281. 11:21of gateway to AI.
282. 11:24Why not just the interface between
283. 11:26people and AI models and you can connect
284. 11:30any AI models to this robot also thanks
285. 11:33to face that has million of AI models
286. 11:38available [music] on the website. So we
287. 11:40can really have the robot as a gateway
288. 11:42for people who are not willing to take
289. 11:45the computer does not or or not able to
290. 11:47take a computer for example senior
291. 11:50people non tech people
292. 11:53so and have the robot as a gateway
293. 11:55[music] to decide given the context
294. 12:00because you can see where is it and you
295. 12:02can know [music] what you're doing what
296. 12:03what people are doing around you can
297. 12:07decide which [music] model to use And
298. 12:09right now for example we have connected
299. 12:11to GP
300. 12:13>> uh to do this uh conversation [music]
301. 12:15demo. So and the robot is aware about
302. 12:19what it can do.
303. 12:20>> Yeah.
304. 12:20>> For example when I ask what what you can
305. 12:22do but you can also take picture and so
306. 12:25on.
307. 12:26>> Mhm. And I think even GPT is not
308. 12:28optimized for this [music] kind of
309. 12:30physical agent.
310. 12:32>> But this agent like rich mini is going
311. 12:35to be the possible [music] next
312. 12:37generation of HCI like human computer
313. 12:40interface.
314. 12:41>> It's it's really how I can see this
315. 12:43become a a big thing like uh like the
316. 12:46smartphone. Smartphone was just an
317. 12:49interface [music] just tactic screen was
318. 12:52the interface for digital world having
319. 12:54access [music] to internet.
320. 12:56>> Yeah. And I think now we are in a new
321. 12:58age where we can interact with machine
322. 13:01using natural languages
323. 13:04>> and not only voice but [music] also
324. 13:06gesture and motion and body language
325. 13:10>> and which is the only way to interact
326. 13:13with within [music] is natural language.
327. 13:16of vision and just and uh and and voice.
328. 13:20[music] And to me it can be if we have a
329. 13:26a lot of uh developers building
330. 13:28application and trying to optimize the
331. 13:31robot can be a [music] way to
332. 13:34to have this as a one of the main
333. 13:37interface to talk to AI. [music]
334. 13:40>> Yeah, it's very inviting to humans to
335. 13:43talk to it. But I can when we have a
336. 13:46different uh chichi or lat [music]
337. 13:49behind it, maybe I can talk to the bird,
338. 13:52>> I can like talk to the dogs.
339. 13:53>> Oh yeah, maybe.
340. 13:54>> So I think it's it's very extensive.
341. 13:58>> Yeah. And and and for example, we we use
342. 14:02it during the week to talk between
343. 14:04[clears throat] me and some of the seed
344. 14:06engineer who are not speaking Chinese.
345. 14:09>> Yeah. So what you really need was the
346. 14:11transcript and it was actually [music]
347. 14:12the first time I was able to speak
348. 14:14directly with a few engineers in Syria
349. 14:17who are not.
350. 14:18>> So uh it can be a more like a a personal
351. 14:22translator instead of tool.
352. 14:24>> Yeah.
353. 14:25>> Can I talk to it now?
354. 14:26>> Yeah.
355. 14:27>> Hi Richie.
356. 14:29>> Hello Richie.
357. 14:35How do I accuse?
358. 14:37>> Uh I and Marshall, we are talking
359. 14:40uh in [music] different languages. Can
360. 14:42you help us a little bit with maybe
361. 14:44French?
362. 14:50[music]
363. 15:00It was a good idea.
364. 15:02>> So he he told us that you can translate
365. 15:05between French and English, but you can
366. 15:06also translate between French and
367. 15:08Chinese. People can choose different
368. 15:10GPT, right? Different light language
369. 15:12model.
370. 15:12>> Yeah, this one was because it was really
371. 15:15convenient to prototype, really easy to
372. 15:17use GPT. And
373. 15:20>> maybe in the future people can have the
374. 15:21local RM with more much [music] more
375. 15:23simplified one.
376. 15:24>> Yeah. Yeah.