Jonathan Victor / Bitrobot
Bitrobot merges AI robots with blockchain through DePIN networks
In this episode
In this episode of Blockchain Interviews, Ashton Addison speaks with Michael Cho and Jonathan Victor of Bitrobot about the convergence of AI, robotics, and blockchain through the DePIN network—a decentralized infrastructure similar to Bittensor, but for robots. Key topics include: ✅ How AI-powered robots integrate with blockchain ✅ The DePIN network and its role in decentralized robotics ✅ Owning and operating your own AI robots ✅ The upcoming Bitrobot whitepaper release
- Bitrobot uses crypto incentives to create a decentralized network of subnets for robotics AI research, similar to Bittensor's architecture.
- Robotics data collection differs from other AI fields because sensor data cannot be easily scraped from the web like text or images.
- Frobots operates RC robots in over 30 cities globally that gamers remotely control while collecting valuable action label data for AI training.
- DePIN networks enable distributed capital formation and coordination among hardware owners, researchers, and data collectors without requiring single corporate ownership.
- The gaming interface allows web2 gamers to contribute training data unknowingly while playing missions that resemble Mario Kart and Pokemon Go mechanics.
Chapters
Transcript
Read the full transcript
I'm ashon Addison from the Crypto Coin Show and today on blockchain interviews with Michael Cho and Jonathan Victor of frobots and bit robot gentlemen welcome to the show and thank you for taking the time thanks for having us thank you very welcome excited to dive into robotics Ai and blockchain has been a common topic on blockchain interviews but
robotics not as much and I'm really interested to see what you two have been working on I would love to just kick off our conversation Michael if you could talk about you know what is bit robot and then we could dive into fro robots and everything about Robotics and AI yeah sure so be robot the goal with this project is really really ambitious I would say so what we really
want to do is use crypto incentives to solve embody AI or basically robotics Ai and the way we intend to do that is create kind of like a network of subnets similar to bit tensor and the idea is that each subnet within bit robot will create or work on specific robotic tasks that are meaningful and will move the needle for robotic research yeah that's great you know
obviously bit tensor tow has had huge Network effect and success and I've seen sort of a dpin networks grow more recently with the launch of some big dpin networks but I haven't seen a lot about Robotics and I feel like that's only the next logical step with AI is putting it into robotics do you think that's the main differentiator which
sets you apart inside of the web 3 space so to say John you want to think that yeah yeah I mean look I think one of the things that's interesting like so I think a lot of times in web 3 we're like Hammer's looking for nails and so it's easy to start from hey look we have this like incredible technology with blockchains and how do we find like the
right problem set to go after I think what's interesting is like if you start from the problem side and then like reason back into like what are the actual things where blockchains might be useful you might find more interesting problem sets to go after so as an example the thing that makes I think like robotics pretty different from other parts of AI is that there isn't
like a giant Corpus of data sets that you can just go after like if you were to say like to build the next robotics like embodied AI model there isn't a ton of sensor data that you can just go pull off the web it's not like for the last 20 years people have been spewing out all the sensor data that you could just go acquire and then run it throw it
into some giant Transformer model and then come out with a model and so this is where it becomes kind of interesting because we're at like this unique moment of both trying to aggregate lots of Hardware in order to create robots of different modalities we're trying to collect data across a bunch of different tasks most of which like require now like also instrumenting
so you can capture the human input as a part of that and then you want to like actually have that be something that you can coordinate across a giant distributed set of people now like in the sort of like web 2 model the thing that we have with like Tesla with figer with some of these like large sort of like robotics Labs is kind of sort of like an endtoend like owning of that
problem where they're both sourcing all the capital to create their own Hardware they're putting that inside of their own sort of like Labs or warehouses so that they can collect their own training data and of getting Loop version of creating that flywheel that we saw with t Tesla and self-driving and I think like this is where if you were asking the question
how do you create like an open equivalent to that like Industrial Pipeline you sort of come to this idea of like hey there's all of these distributed parties that you need to coordinate and that can go from people who are capital owners so people who might want to contribute robots into a giant Fleet that could be researchers that maybe don't have access to billions
of dollars of Hardware to go evaluate their models models or run training tasks or even acquire the data sets or Bounty out tasks that might be useful to input into future models and so on and so I think like crypto becomes an interesting solution to this problem because you can take the constituent things that you need in that Industrial Pipeline to create a final embodied AI
model and say you don't have to own that entire pipeline by yourself as a corporate Lab in order to create a solution you can actually say you as individual participants in the supply chain can be mind around a shared economy where you are getting commerate like rewards for your contribution inside of that but we can create the full and 10 value Loop that rewards
everyone along the way and I think to me that's where like deepen and like Capital formation is like quite interesting because robotics actually is quite Capital intensive given like we're trying to build like physical infrastructure but also it gives you this way of saying like how can I like Focus people on specific tasks and use sort of like the subnet model
specifically as like an incentive to like really focus on I need robotic arms that will F fold laundry so I can collect data on that problem or I will do Urban navigation so I can collect data on that problem and so on so sorry that might have been a little too longwinded but I think like really pointing out I think dpen has a really unique opportunity both from like the
capital formation side because robotics specifically has like a lot of capital needs but also on the coordination like take all these disperate like parties that are operating in this robotics pipe and coordinate them together yeah no Jonathan that's a great explanation and showing the difference between the Tesla Model and how do we fit that into web 3 or in this case more
of a dpin network decentralized physical infrastructure Network so could you explain that a little bit more in how do contributions work if you're looking you know if you're a deepin robotics platform and you're looking to gather the hardware or the software or the input information to build this how do you set that up and then get people
to start contributing into it Michael do you want to take this because I feel like me yeah so maybe it's it's let let me give some very concrete examples and maybe tie back to like what you know I've been doing at photots the last nearly 3 years now so we started actually building this Salk robots and I actually have one of them this is our latest version
It's like a toy it's like RC RC car you put in a 4G SIM card and you can actually remote control it in away from anywhere cuz the whole thing it's it's it's what they call T operated so you know you could be sitting in New York and yet have this robot you know driving around in somewhere in Africa and so we actually structure the whole thing as a robotic game when I
say game I literally mean like a video game so think of it the gaming experience is kind of like Mario Kart SL pookemon go so you'll be you know as a gamer you'll be looking through the camera of this s robot and then we give you missions you it's a little bit like Hunger Games where you go around and then you pick up some loot boxes and pick up Solana nft and all
that ja right so except that of course it happens in the real world so to The Gamers it feels like a real game and actually we do have Gamers who actually play this and in fact mostly web to Gamers they don't even know this there's a crypto angle to all this but really that is a means to an end so as The Gamers play the game we actually collect data right
anything from the you know the video stream coming from the cameras but actually more importantly from you know what the what we call the action labels you know what the gamer human gamer is doing on it and it turns out that data set is very rare like what John was saying is like no one actually actively do something like this for I would say we are probably one of the
first team in the world to try to even attempt to do this and it's very rare data set but it's also act to be very valuable for robotic research where let's say you and in fact the pipeline is very similar to you know what Tesla is trying to do right so Tesla will you know gather video from the cars but also they gather what
the human driver is doing again action labels and so we just started working on this for the last 2 three years now we have Bots you know over 30 cities you know Africa South America you know North America Asia and whatnot and so and then we started actually working talking to some researchers and to kind of validate
the thesis hey maybe this is actually useful and I'll say fast forward today today like that there's a lot of validation on the front so for example last year we co-organized this kind of AI versus human competition ition with a bunch of researchers from Deep Mind we're pretty close to the guys there now and we got a lot of really great research Labs taking
part and you know now we have like different research collaboration with Labs from y berley Stanford you know and so you like institutes like that and so I think it's a it's a great proof point that you know you know despite us being such a small project and in fact despite us being crypto CU usually the researchers are very you know they think mostly crypto is
scammed right and despite all those they actually get to work with these really worldclass researchers so I think it's a big validation that indeed this data set is on to something and that's how we started right you can but you can think of what we did the last two three years as literally just subnet one and I'm talking about the game and in fact the
game the competition that we actually Co organize with the Deep Mind guys at subnet too because if you look at subnet one we are trying to incentivize human Gamers and possibly people who you know buy this robot by the way this like this thing cost like $150 so it's a very deepin you know you know friendly price point let's say so that could be subnet one right so that's
where you're trying to contrib attract like incentivize these two grow people and you look at subnet two which actually could just be a much bigger version of that competition AI versus human kind of competition that we had this is where you actually trying to incentivize researchers and in inde you actually also need to attract compute providers
cuz it turns out the data set that we actually give to the researchers they literally need you know number of h100 to even run the model and so but you look at subet two the output is actually at AI model whereas subet one is a data set so and if you think of it this way this can be extended Way Beyond sidewalk robot it could be for humanoids it could be even for things that it's
let's say in simulation right and B over there you also need to incend researchers you need incentivize again GPU providers to do all that and as John say all these are very Capital incens intensive and in some cases you need a lot of actually you know manual labor to do all this things and so I think crypto just have this unique advantage to do Capital formation
globally you know very quickly and yeah so super excited to actually move Beyond sidewalk robots to all all these other types of robotic tasks Definitely Maybe another little settle point here is like robotics is actually I mean so there's really cool demos that I think a lot of us see on Twitter every now and again but like there's a lot of large
unsolved problems so the reason that Elon talks about like optimus's hand so much is like manipulation is like one of the hardest problems and so I think this is actually where like web 3 is actually not that far behind web two and in fact like outside of things like simulation I think there is actually like a real opportunity to dramatically scale up how much like human sort of
like labeled data can we acquire and that's something that is literally just a human throughput problem which crypto has shown an ability to do in parallel quite fast compared to like web two so I think this is actually where like it's not like we have a real shot on goal I think for like making web 3 like the foundational platform for a lot of this
stuff that's very exciting and Michael that was a great little show there of that RC car I'm picturing like from kids playing with remot cars in their backyards to moving into you know controlling a Rover on another planet or something you know if you can control it in another continent another continent I think that's the next step so that's
really cool and I love the gamified aspect of it I feel like you know when when Pokemon go and Ingress which I played before Pokemon go which was the basis for it the software you know I that went viral so to get the physical infrastructure in there and then add robotics in and use that to train the models I feel like is a great way to kick off building
those data sets and you know John I'm curious if you can sort of paint that picture from Gathering the actions and the models from the RC car and the gamification in the first Subnet how do you take that information and use that to build a long-term vision of having robots that are you know in our house or on the street doing things that people see hey this is actually changing
the world or making my life easier yeah so I want to like maybe set expectations properly the analogy I've used quite often is like whenever you look at these like really big types of innovation it's sort of like the integral of many S curves together so I don't think you should expect like overnight you go from Like A to B what you should see is like
if you're looking at the final output there's a bunch of input progress that to happen and the right way to look at this is like how are those things improving because like as you hit more and more of those improvements you'll then see like that s the macro S curve really show up and it would look like that and so I think specifically the thing to keep an eye out for is just how
are we like looking at specific tasks so I think like when you talk to researchers and I think a lot of sort of like why we spend so much time talking to researchers is to figure out like what are the actually like most important proximate problems that don't have a clear line of sight to their Solutions so that you can prioritize which are the tasks that you want to go
collect data for that gives you the ability then to actually like figure out what are the biggest bottlenecks that like web 3 can actually go after to help solve the problem faster so like Urban navigation is definitely one I think this is why we've been talking about robotics arms and like manipulation is like another important one but I think that's how you sort
of prioritize there's a great paper that came out I forgot the name Michael can definitely fill it in on cross training and I think it's like there's evidence that shows that when you have these like embodied AI models that are trained on top of data sets that came from sort of like heterogeneous sources the combined model is actually better than like if you were to just singularly
focus on any individual task so I think there's like also pretty good research that underpins this idea that look if we can collect data from all of these different sources on all of these different tasks the combination of that will create a better model as well which I think is like part of why we're like pretty bullish on this approach of like multiple subnets focused on different
tasks that are really like trying to prioritize what are the biggest gaps inside of like the robotics landscape MH definitely and you're right it's not something that can happen overnight but I feel like the pace at which AI is improving is faster than people thought you know I seeing the different versions of just the llms coming out each year it's getting
like 10 times or more better each version how quickly do you think that the robotic side will will grow in a sense that real world people will see how much functionality has been improved in a short amount of time yeah so maybe I'll take this one so yeah I think that's a very good question so because that's this word AI in embod AI so people start to come
with this expectation that you know the velosity or the iteration cycle is going to be as quick as you know things like the latest ROM right but if you think of all this AI development in the digital RM or the cognitive R so really the only three things that matters is you need data right and then you need compute and you need
talent right mostly researchers that can push the boundary of the latest algorithm right and you basically need only these three core ingredients and so for example when let's say deeps or like open drop another model what you realize is that you know typically within half a day or one day right in AI Twitter you know land pretty much everyone will have a pretty good idea
how good or how bad is that particular model right even though there are a bunch of different benchmarks but in general you can arrive at a pretty good idea how good or how bad is that model meaning the last step of evaluation or benchmarking is never a bottleneck right so now if you think about embod AI it's very different so let's say when Tesla rote out their
latest like FSD they have no idea okay let's say within one day okay and let's say they've already collected 1 million miles and there's no crashes or no disengagement in the in the 1 million miles so the question is have they actually like reached level five or are they still a level two thing because what if in the next MTH there's a crash and so if you think about it like that
right actually in robotics and the evaluation step could actually be a way better bottleneck even compared to the lack of data right we kept talking about the lack of data just now and that's a serious bottleneck but if you think about that actually evaluation might be an even bigger bottleneck because you can't tell how good or how bad is your model in
simulation you actually need to do it in the physical world where the robots Are Meant to do the physical things and so I think because of this particular like evaluation step I think a lot of people are going to be rather disappointed that it's it's going to take a lot longer you know than expected but you know I think in if you take that stance then actually
perhaps robotics a or embody AI might be the final you know Frontier left after we've we've gone past like all the cognitive task that most humans can do it could be the final piece in the path to AGI and so yeah I think all this just B to me it just means that like in the coming years there just going to be more attention and more like
funding looking into like solving this robotics issue and again you need a lot of capital you need lot lots of physical robots to actually test this these things out in the real world which actually maybe to like add on to my point I think this is actually like a really important like piece of nuance so when you actually think so let's take cars as an example like you can have a
car that's useful without self-driving and so you can distribute the capital base to customers to like pay $50,000 for a Tesla or whatever and like Tesla can run and all of like their data Gathering evaluation like in Shadow mode on the car like after the fact and they've basically gotten like lots of capital support to basically build a self-driving Network the same is not
true for Robotics and like that's why like for a lot of the robotics like you need to like start with something where if the robot is going to cost5 to $50,000 to $100,000 it's a much more Niche set of things if it doesn't actually perform ATT already and so that's why I think you see a lot of like the early sort of like boot straic of this being like
Tesla fig like people doing things inside of their own like warehouses where they can control sort of the environment they can build it up to you like some sort of utility locally for themselves and like Michael has made this argument to me before about how like sort of B2B is probably the first place you'll see this where like for very specific tasks you will see robots
that are like hyper trained to do those things but on the evaluation point I think this is actually where like the argument I was making about those S curves is really important because like evaluation is sort of like the last part of that and this is where dpin could change where like how fast does robotics come because if we can get a giant distributed base of robots that
are available for evaluation you've now increased everyone's cycle time in order to evaluate their models so like that final sort of like thing can move up but yeah so I think like that's one area where like we can make a prediction about this but a lot of it is sort of path dependent if there are 100 or like a million robots that are available for
parallel evaluation you have like a million x the ability for people to do their evaluation Stu which actually makes it much faster to like do the iterations to like get to the next level yeah and maybe just to Circle back like what we were discussing using gaming as a as the wedge right and I think makes a lot of sense at least to me because I think that I take the view
that actually for a robot let's say a human to you know to actually take over a chef at a restaurant in my view is at least a decade away right and so in the meantime what do these robots do if it's it's it's just going to have like cuz it's it's I think for most use cases is not going to be good enough 90 if the robot can do that task 99%
success rate what you actually need is probably Six Sigma right 99.9999% right and you we've seen this in self driving I remember sitting in I got a chance to sit in a Google self driving car back in 2015 this is 10 years ago and I thought wow okay this is coming and of course now we're here we're still talking about you know in going to production so these things
might take a long time and so I think in the meantime actually gaming is a is a very I don't know whether you can call it a use case but you know it's certainly have a lot of entertainment value and so and often times if you look at gaming it has been that Genesis of pushing the new wave of Technology you know forward it could be in fact the
fact that we have GPU today doing AI is because we have GPU running for games for decades right for NVIDIA to back build to this point or VR and all that so yeah so I think I think gaming is going to be U you know I myself I'm I'm I mean i' I've driven out robots you know in the real well I just find it very fun and so it is useful in that way
to me already and that could be the wedge that we need actually to tie things over while we collect data in the real world definitely yeah I think that personal robotics is it's so specific in the tasks that a robot needs to complete that I think people are underestimating just how precise it needs to be it's not just like a robotic arm that just does one thing it really
needs to do everything that humans can do with such Precision that there's there's still a long way to go so I'm I'll check back in 10 years with you and see where we're at yeah now with the contributing into this robotics movement and the and the dpin network specifically is there is the main way to do it right now just getting the
the RC car and playing the game or are there other ways of just contributing your GPU power or finding other ways to input software or information into the AI model yeah so I'll take this one first so obviously we just announced the concept of bit robot and the Heart paper it's it's going to be published in you know a couple of weeks time and so you know I think in the
coming months we will have more concrete proposal or like you know setting up the actual Network and going into test net and all that by the way this B robot will be built on Solana so I think officially like in a couple of months time you see concrete like ways for people to participate having said that I think in the coming weeks in fact we
have actually you know a number of really interesting in my view partnership with actually other dpin projects where these are I would say all these partnership are you know early Showcase of what you know future subnets could be and so you know this would be you know where is kind of similar to the things that we've been illustrating just now this could be you
know certain projects that have a lot of compute GPU this could be projects where like like existing project like ourself you know the with the side robots and it could be projects that actually have a lot of let's say you know a community of Gamers right so those SKS in fact we are close partner with Y Yu games so there'll be projects
Partnerships like this where we'll continue to Showcase not just of side robot like I said in the coming weeks or months and so yeah I think I think a lot of this will be a lot more Concrete in everyone's eyes in once you know all these partnership starts showing up and of course not forgetting that we have you know pretty good relationship with a bunch of
research labs and a lot of these partnership also involve those labs because honestly without these researchers even if you provide a computer like who's going to run the GPU right you know so we really need those Talent to be involved as well definitely yeah I'm looking forward to seeing the paper and I know it's early on and you know growing the aiel
models is one thing but actually putting it into physical robotics is the next level which is it's a lot of work and I'm I'm glad that you guys have these Partnerships with these research Labs but because without that it would be even tougher maybe you can give a quick 30 seconds on the rationale or the reasons of why the salana chain is good for a deepen Network and for
building out this robotics platform yeah so I would say so when we've started like three years ago honestly I was very anti- crypto so but what happened was I discovered like I think me and my brother who's my co-founder so we discovered helium like two months into the project so at the time before
discovering helium we were just thinking of okay creating a sidewalk robot that's a spre sh so that was our thesis and then but two weeks later sorry two months later we discovered hel and I was just mind blown right cuz you know they didn't they didn't have this term called deepin back then but I already saw that hey you know they Meed a
token out of thin air and they were able to basically cross off so much capital and what is even more impressive is that they done it so like on a geographically distributed Manner and so shortly after that I was like okay I need to learn how to do this and then literally one week after that there was a Solana hackathon in Singapore and so we didn't know what we were doing you
know I didn't know what is a smart contract what is a L1 but like and so but yeah gosh this is like nearly 3 years ago and so for us in FL Bots we've been actually even though we don't have a token you know but we've been messing around with a lot of Solana nft and all that so we've been very comfortable actually building on it you know I would say what we do 99% of
it like doesn't like the success is depending on how good is our Hardware you know how good is our networking stack and things like that but for us as a builder I think you know Solana has been a decent place to build on and of course like you know it has survived you know you know many setbacks and nowadays you know you have very very big deeping projects you know for
example helian B to Sol so I think that giv us a lot of confidence that this is a decent you know L1 chain to build on and of course like you know with the recent round of funding we were really lucky and fortunate to have backings from Solan Ventures and you know Angel checks from anatol and rash and so yeah so we are definitely I guess Solana boo y yeah
that's great no that's that's wonderful to hear I'm looking forward to it and so now with a future sort of tentative road map which of course everything can move around depending on how quickly or the quality of the AI information that's gathered but you mentioned there's a white paper coming out shortly and then there's also the salana nfts that you mentioned
interested in is that something that when when can people start accumulating that and then what else can they do in the near future as the iterations keep growing yeah so I'll would say the nfts these are not like the you know 10,000 limited version these are literally just like in-game assets so like you shouldn't think about it's just
yeah we use NF like for the utility purpose not not like yeah but we actually have this other thing called Sam so Sam is actually an embod AI agent with his own mem coin and if you go to the handle Twitter handle is called Sam is moving and it's basically live streaming 24/7 so anytime you look into it you see an AI That's driving one of our robots you know in a
random city and literally the whole thing is 100% right AI you know autonomous driven and so yeah and Sam Sam is actually built on virtuals so and yeah in fact we are yeah we have some Prett big announcement I mention that virtual invested in know in know this current as well so we're very close to them and so I'll say for people who who are
interested to accumulate something but no not Financial advice I actually also mentioned this is just a mem coin right no pretend of utility but I think you know for me because I again I came from 3 years ago completely thinking the crypto is all right and so it took me a while to understand what meme coins represent and I've come to realize that
actually you know of course on one hand I'll say our north side is definitely what we are you know building with Jonathan and the protocol laps guys like very de you know heavy Capital formation so that's one extreme of crypto you know if done well it can really move the needle and that's truly on off star about but the other extreme of crypto which is all the DJ all the
mcoin right you know you have really fickle-minded like let's say people a community but then again I think it's also a great place to actually very quickly Gana attention maybe at Capital as well so I'll say I mean in yeah in very very soon we will have some announcement on that front as well but yeah do check out Sam you know it's it's it's a virtual it's a
pair on virtuals so very cool yeah I'll add the Sam's X handle in the show notes as well I'll check that out on virtuals and great to see they're partnering with you guys Jonathan do you have any last words on you know the plan moving forward and how else people can get involved before we wrap up yeah I mean I think for right now
maybe the thought would be try to follow on the different social channels so maybe we can give you like the bit robot Twitter handle and things like that so people can say up to date I think in the coming like weeks to months obviously like white paper I think as Michael mentioned a number of like Partnerships with other deepin projects and then once the network
launches like we have pretty ambitious targets for how many subnets we hope to spin up so I think once those start getting announced there will be a bunch of different ways for people to participate and help contribute whatever sort of resources they have whether that be their own time or yeah buying robots and things like that very cool I'm looking forward to it
Can't wait to see the robots driving down my side street here whether it's me it or somebody else in Singapore or another continent which is even cooler I really appreciate you guys insight into where we are at with robotics and AI I have a feeling like more and more people are going to become interested in it from the blockchain
realm and AI realm as we get more physical here and it looks like you guys are really spearheading something cool and I love the gaming aspect at the beginning I think it's a great play looking forward to the launch I'll leave the all those links you mentioned in the show notes and would love to guys have you guys back on after the white
paper is launched and the subnets are are going live to see how else people can contribute and so I can learn more about the best way to get involved and participate and I really do thank both of you guys for your insights and your time today thank you well yeah thanks for having us
More interviews
0G Labs' Michael Heinrich on cheaper open AI modelsNov 19, 2026
Why Wall Street is moving toward 24/7 trading in 2026Nov 13, 2026
How Wallet V is letting AI agents trade while keeping your keysNov 10, 2026
How Silent Payments Keep Your Bitcoin PrivateNov 9, 2026
How XYO is building proof in a world of AI fakesNov 5, 2026
How Cap financed a Hollywood film with onchain creditNov 4, 2026