Why I Spent $199 on a Sidewalk Robot Instead of Another Memecoin
What starts as a game ends up collecting the messy data that could power the future of robotics
Introduction
Remember when Pokémon Go took over the world? Everyone was glued to their phones, wandering through parks, chasing virtual creatures, nearly walking into traffic. It was chaotic, but also kind of magical. More importantly, it proved something big: blending the digital and physical world could actually move people, literally. And it wasn’t just hype. Your friends, your cousins, even your work group chat were in on it.
Lately, I’ve been getting that same spark again, but this time, it’s not about catching Pikachu. It’s about watching sidewalk robots stroll through real streets.
Meet the Earth Rover Mini+, a $199 browser-controlled robot from FrodoBots that you can drive from anywhere in the world. And before you roll your eyes: no, this isn’t some overpriced “crypto toy.” It’s something else entirely.
I went with the Mini+ over the $149 base model mostly for the upgraded camera (more visibility = smoother driving) and access to the Earth Rover Challenge, where humans and AI agents compete in real-time missions across real-world terrain.
Imagine this: you’re playing a game where you navigate a robot through real-world streets: dodging pedestrians, clearing checkpoints, dealing with live internet latency, and every move you make is generating the messy data that robotics desperately needs. This isn’t just remote control for fun. It’s AI training disguised as gameplay.
I’ve been deep in Web3 gaming since the early days, so this checked a lot of boxes. FrodoBots is making teleoperation fun, affordable, and accessible. It’s a new take on embodied AI: one that feels more grounded, and frankly, more playable. The setup is simple. All you need is a SIM card, decent internet, and a browser. Then you’re live, piloting a real robot through the streets of some faraway city.
Now, I don’t use the word investment lightly anymore. We all know how that plays out in crypto. But this wasn’t a degen move. It was driven by curiosity, passion, and a lifelong obsession with RC cars. The best part is this isn’t just FrodoBots shipping bots. They’re building a game that teaches robots how to handle the real world. And maybe, just maybe, you get to earn something along the way.
Well, I’m still waiting for my Mini+ to arrive. But that didn’t stop me from diving into the project behind it.
And the more I dug in, the more serious it got.
They recently published a research paper titled “Autonomous Mario Kart in the Wild,” co-authored by researchers from DeepMind, UT Austin, NUS, and Seoul National University, many of whom participated in FrodoBots’ Earth Rover Challenge in Abu Dhabi in 2024.
The paper explored lessons learned from “in the wild” evaluations using FrodoBots’ hardware. It’s a strong signal that their work is more than just experimental, it’s gaining the attention of some of the sharpest minds in robotics and AI who saw enough value to turn those insights into peer-reviewed research.
That momentum extends beyond the challenge itself. Even Sergey Levine, one of the top minds in embodied AI and now cofounder of Physical Intelligence, credited FrodoBots for providing the hardware and infrastructure that powered their recent work. Using FrodoBots’ platform, they trained a robotic foundation model that could drive mobile robots across six different countries.
To some, it might seem like a random impulse buy. But for me, it scratches a very specific itch: part childhood dream, part curiosity about how gaming might actually help solve one of robotics’ biggest challenges: collecting messy, diverse, real-world data. And I genuinely believe this playful, game-driven approach could unlock something new for physical AI.
So while the robot’s still en route, I’ve already gone down the rabbit hole. Here’s what I’ve found.
The Bigger Picture: What Problem Is FrodoBots Solving?
FrodoBots started with a simple idea: let people control sidewalk robots through a browser, collect real-world data, GPS, video, audio, and reward them with crypto. The goal? Build a DePIN (decentralized physical infrastructure network) where anyone can help train smarter robots just by playing.
Co-founder Michael Chung Yeung Cho later revealed a much bigger goal: building the BitRobot Network, a modular, on-chain system designed to “solve embodied AI.” FrodoBots will become just one subnet of this larger vision, feeding constant, high-quality data into a broader network of robots working together to power the next phase of physical AI.
Why does this matter? The answer: embodied AI bottleneck.
While today’s large AI models are trained on trillions of tokens scraped from decades of internet content, robotics doesn’t have that advantage. Physical AI needs real-world interaction, moving through space, avoiding obstacles, and reacting to unpredictable environments. That kind of data is slow, expensive, and difficult to gather. Outside of a few giants like Tesla, most robotics teams are starting nearly from zero.
Unlike text-based AI (which has massive training sets), physical AI doesn’t have a rich dataset of real-world navigation challenges, which are full of complexity and unpredictability. This is the gap FrodoBots is trying to fill. Their whitepaper lays out a vision: a global swarm of low-cost, user-controlled robots collecting rich data from real environments. Each mission, each bot, each run helps train better, safer AI.
And that leads us to the core of the experience: the EarthRover.
EarthRover Sidewalk Robots
EarthRover is FrodoBots’ browser-controllable robot built for urban terrain and real-world navigation, that can be piloted remotely from anywhere in the world. Unlike traditional RC toys, this bot is designed with purpose: to generate valuable raw data from real environments. Think of it as a research vehicle disguised as a toy.
What makes it unique is its accessibility. By rethinking robots as toys, they drastically lower the production cost of each robot unit, making robot ownership affordable and opening up a new kind of community where gamers, hobbyists, and researchers alike can contribute to the future of embodied AI. It’s robotics for the rest of us.
Earlier this year, FrodoBots took things up a notch with the EarthRover Challenge, a wild experiment that dropped sidewalk robots into the chaos of real cities and let humans and AI agents battle it out in live navigation missions.
We’re talking about bots rolling through the actual streets of Kenya, Costa Rica, Mexico, Brazil, and Argentina, not test tracks or simulations. Real sidewalks, real pedestrians, real GPS glitches, and the kind of unpredictable mess that makes robotics hard (and interesting).
Each mission came with a time limit and a series of GPS checkpoints to hit. Human players and AI teams both gave it a shot. The results were interesting as some cities were friendlier than others. But the real win was in the data, every run, win or fail, helped build FrodoBots’ growing dataset of edge-case navigation moments. That’s gold if you’re training physical AI.
They’d done something similar in Abu Dhabi in 2024, but this time it felt bigger. More global. And with each event, it’s becoming clearer: FrodoBots isn’t just shipping cute robots, they’re building a playable infrastructure for robotics training.
The best part is learning about the fun ways you could earn with these robots. But we’ll get to that.
Robots That Compete, Earn, and Rent Themselves Out
FrodoBots isn’t just about driving a robot for fun. It’s quietly a game-powered earning network where your EarthRover can work for you. Here’s how it plays out: when you’re not using your bot, you can rent it out to other players or AI agents. If your robot completes missions or wins challenges, you get rewarded.
A perfect example is SAM (Small Autonomous Motherf*cker). He's an AI agent launched through the Virtuals platform who competed in the 2024 EarthRover Challenge. No extra tuning or human help. Just the baseline model navigating real-world streets and still pulling in a solid 5.86 points. Now, imagine SAM renting a whole fleet of EarthRovers owned by users like you. She completes tasks, earns rewards, and everyone in that loop gets allocation.
This vision is already being tested through robots.fun, a daily game arena where bots: human- or AI-driven, chase down alien NFTs in a two-round bounty hunt. One round on their own bot. One round teleporting into someone else’s. The top-scoring agent wins the day, and SAM buys $1,000 worth of their token, holding it in her wallet.
And it’s just the beginning. FrodoBots is laying down the groundwork for a full robot-agent economy, with more features in the pipeline:
Custom mission creation and hosting
Dynamic rental pricing for bots
Token rewards for bot owners, drivers, and agent hosts
Revenue sharing as the network (and its dataset) grows
It’s early, but the idea is clear: your robot becomes a real-world game piece, and an income-generating asset. You don’t need to be a robotics engineer or an AI dev. You just need an EarthRover, a browser, and a bit of curiosity.
It’s like Uber, Pokémon, and passive income rolled into one. The thought of your RC bot cruising through a town on the other side of the world, catching digital aliens or collecting sidewalk footage? Yeah, that’s the future we’re walking into.
Still not sold? Watch this robot buy lemons from a sari-sari store in the Philippines, all remotely controlled through a browser.
Why This Actually Matters
While most of the tech world is busy doomscrolling AGI timelines and debating whether ChatGPT has feelings, something far more grounded, and arguably more urgent, is happening on sidewalks.
This is the first time anyone has figured out how to crowdsource embodied AI training data at scale, using consumer robots, and make it profitable for participants. That’s a huge unlock.
Because here’s the truth: robots still struggle with the real world. Despite billions in AI funding, physical AI, robots that move, sense, and react in the real world hasn’t kept up. Not because of bad hardware, but because of bad data. You can’t teach a bot to cross a street using datasets scraped from Reddit posts.
RC Cars Walked So FrodoBots Could Run
Every now and then, we get the chance to fulfill a childhood dream, sometimes it’s through travel, food, or finally owning that one thing we always wanted. For me, it’s this: a remote-controlled robot that can navigate real streets, stream data live, and maybe even earn crypto while it’s at it.
FrodoBots is a glimpse of how robotics, AI, and gaming could merge in ways that are both fun and foundational. And once my EarthRover Mini+ lands, I’ll be back with a follow-up.
Until then, the sidewalk awaits.
About Conglomerate
Conglomerate is a seasoned content writer and KOL in the crypto x AI x robotics space. Web3 gaming analyst, core contributor at The Core Loop, and pioneer of the onchain gaming hub and Crypto AI Resource Hub.
Book A Call
Curious how robotics, gaming, and AI can drive your next growth wave? Let’s talk. Book a call with Crescendo’s BD lead Filippo