Harry O'Connor
226 posts
Harry O'Connor
@Harryoc493
Harry | Robot Chiropractor & Researcher |
Ca Katılım Aralık 2018
2.7K Takip Edilen1.4K Takipçiler
@Sentdex @carlosdponx @chrisbolas Agreed, their hands are ideal. I think this style of the dex-3 style is the best of both worlds in terms of dexterity and robustness
English
@carlosdponx @Harryoc493 @chrisbolas I also don't value super nice hands, but I do want hands and enjoy that problem space. I just want them to have basic pick up and drop capabilities. Truly dont care if they can thread a needle. I just want them to grab stuff. I think sprouts hands look good. Same w/ Sunday
English
Wow, they really nailed the aesthetics. Would love to try deving on this.
2dof head
prominent estop, wonder if also a remote estop.
no pinchpoints.
50lbs.
Appears to come with hands/dev friendly @ $50K.
NYC based.
Definitely one to watch.
Fauna Robotics@faunarobotics
Meet Sprout. Today, we’re releasing a new kind of robotics platform. One designed to move out of the lab and into the real world, closer to the people who will shape what robots become next. [1/4]
English
Yeah it is a problem lol
I think the hands issue is a general thing to be solved.
When I was at ces I went to every gripper booth and asked “Every hand I’ve tried breaks after one impact have you done durability testing?”
Every time I got the same answer:
“no, no haha it’s much too early for this. BUT ITS GOT TACTILE SENSORS!!!!”
….
English
@Harryoc493 @chrisbolas @carlosdponx I do love my G1, but the G1 falls so damn hard, and the hands do NOT survive real usage, despite being absurdly expensive. When stuff does break on the G1, getting parts is a nightmare across the ocean. I do not think the H2 is coming with hands @ $50K, but that'd be interesting
English
Its a really nice design - it being soft and having no pinch points is awesome
With that said. a H2 from Unitree EDU is $50k - that's 6' tall, (with hands? afair), 3/4 hours of battery life?, so its capable of essentially almost any task in house like counters and shelfs
I really, really like the sprout design, I'd be willing to pay ~85k for a taller version so it could do shelf/countertop work, this seems a little sall especially for 50k, if it was like $20k I'd insta buy. The Kbot was an ideal size for something like this imo.
English
English
@Sentdex @carlosdponx A g1 slaughters both in terms of cost : performance
Also a big + for the g1 since there are like 10k papers you can implement on it day one at this stage. None for the booster
This bot looks so sick, it just needs to be 1/2 the price.
English
@carlosdponx thats 3.5' tall SIR!
You can get a fully upgraded G1, with hands, maybe for apx that price, yes, but only if you're riding on a bulk order. Otherwise you're paying more.
You cannot get 10 boosters w/ hands for this price. The comparable Booster is apx this price afaik, ~42K?
English
@Harryoc493 Thanks! Not publicly but we have done a *ton* of market research and the pricing will surprise our industrial customers in a good way.
English
Excited to build a future where robots improve our everyday lives youtu.be/sd8ivhpjI6g?si…
YouTube
English
@KyleMorgenstein I actually agree with all of this. We should chat offline sometime, I think you'd find my approach quite interesting.
(removed a lot of the pen's/rewards so I could iterate faster to figure out what causing the imu issue with DR)
English
@Harryoc493 I agree that you don’t want to be overly prescriptive with the rewards, but otherwise just do velocity tracking + min energy, but in practice that doesn’t work
English
@KyleMorgenstein I largely agree, generally I try to avoid specific gait enforcement as much as possible. Imo this results in essentially a sub-optimal variant of expert cloning
(this policy is dogshit - 200 iters. It's a very basic rew structure today while I debug imu issues on s2r)
English
@Harryoc493 there’s a spectrum, some are overly restrictive (all the phase based ones) but without them you often get behavior like this. I usually add them formulated around air time symmetry rather than around explicit gait timing which is much more flexible imo
English
@KyleMorgenstein I dont have an airtime reward or foot contact reward.
I had that issue when I tried the g1 env a few months ago lol. This is why I largely dont like the hand crafted gait reward styles.
English
@Harryoc493 usually this indicates a problem with the gait/airtime reward
English
@tbpn @t_xu @stanleytang @Sirupsen @zcabrams @andrewdfeldman @BChillman @alexalbert__ @arthurquerou wen @bentay 👀?
x.com/benktaytay/sta…
Ben Taylor@bentay
Your Mac just got a brain. 🧠 Spec proactively drafts replies, summarizes docs, and helps before you ask. It connects knowledge and memory across iMessage, Slack, email, and your calendar, finally making your apps work together. Goodbye silos and context switching.
Morning. On today’s show:
– Agentic commerce is here
– OpenAI drops Sora 2
– @t_xu (DoorDash)
– @stanleytang (DoorDash)
– @Sirupsen (Turbopuffer)
– @zcabrams (Bridge)
– @andrewdfeldman (Cerebras)
– @BChillman (Phantom)
– @alexalbert__ (Anthropic)
– @arthurquerou (Vibe)
Tune in!
TBPN@tbpn
Tuesday, September 30th x.com/i/broadcasts/1…
English
Your Mac just got a brain. 🧠
Spec proactively drafts replies, summarizes docs, and helps before you ask. It connects knowledge and memory across iMessage, Slack, email, and your calendar, finally making your apps work together.
Goodbye silos and context switching.
English
@campedersen Not at all it’s actually dragging it (and me lol) here a bit. You can see towards the end.
It’s trained with and without a winch/stuff pulling it. I’m not finished training - that was an early test so keeping it on the winch for a while longer.
English
An electric horse you say?
Dave's trying to bring me to his lair
(100% RL, sim2real 60m of training)
English
English
@eigentime5454 @chris_j_paxton no I don’t think it’s better but I think it’s good enough to not outweigh the other features it has. both Mujoco and Isaac will have mj warp backends so there won’t be any difference soon on that front.
English
Isaac lab is powering basically everyone right now
Ankur Handa@ankurhandos
Our whitepaper on Isaac Lab is out! Isaac Lab is a natural successor of Isaac Gym that pioneered GPU-accelerated simulation for robotics. It subsumes all the features of Gym and provides the latest advances in simulation technology to robotics researchers. It also supports warp-based custom sensors, actuator models, motion generation pipelines, teleoperation devices, and various ready to use environments for sim-to-real research for locomotion, manipulation, navigation and more.
English
Indescribable feeling to see a humanoid startup on my timeline using a K-Scale robot. Makes the past two years of eating glass feel worth it.
Harry O'Connor@Harryoc493
An electric horse you say? Dave's trying to bring me to his lair (100% RL, sim2real 60m of training)
English