The world's open robotics lab. Apply now.
bitrobot.ai
- Tweets 1,531
- Following 1,749
- Followers 114,744
- Likes 4,433
Pinned Tweet
1/ Better robot models come from better data.
That takes two things: hardware that captures it fully, and tooling that turns it into training signal.
We built the first. Today, @GroundedSI launches the second: Grounded API. Access both hardware + API ↓
33
74
657
240,862
Starting next week, the Earth Rover Grand Challenges will be at @ieeeiros IROS 2026
11 teams will test their navigation policies in 4 tracks ranging from city to indoor / off-terrain to marathon (50 miles of sidewalks, overpasses and bike paths)
Let the competitions begin 🦾
Nobody has driven a mini rover from UC Berkeley to Stanford on its own.
Humans have done it by remote control, navigating 50 miles in 29 hours and 38 minutes.
That's the benchmark to beat at the Earth Rover Challenge co-organized by @BitRobotNetwork at @ieeeiros IROS 2026 ↓
4
2
14
1,493
BitRobot 🦾 retweeted
Packed room at our office launch last Friday!
Builders from Physical Intelligence, Google DeepMind, Tesla, Anthropic, Dyna + beyond explored the full human data stack we’re building, from capture hardware to enrichment to training-ready outputs.
GSI was built to plug easily into the training flows teams already use.
4
8
38
6,837
Excited to co-organize the Origami Grand Challenge with @SharpaRobotics at @ieeeras IROS next week.
Sharpa builds advanced dexterous robotic hands and embodied AI models.
Over a dozen teams will put their policies to the test by autonomously folding a paper airplane.
Sharpa will be co-hosting two competitions at IROS 2026.
Powered by Sharpa's robot platforms and software, the 2nd ROCO and Origami challenges (cohosted by @BitRobotNetwork) brought together 48 teams from leading institutions around the world—including Carnegie Mellon University, UC Berkeley, KAIST and NUS.
Across both competitions, teams will put robotic manipulation and dexterity to the test on Sharpa hardware through challenging real-world tasks.
Follow the challenges for more:
ROCO: @NtuLab38456
Origami: robotic-origami-challenge.gi…
6
8
35
3,454
Next week, we’re heading to @ieee_ras_icra IROS 2026 in Pittsburgh with 50+ research teams pushing the frontier in 3 Grand Challenges:
> Autonomous navigation w/ Earth Rover
> Dexterity thru origami folding
> Loco-manipulation thru Ikea assembly
Excited to organize this w/ partners like @SharpaRobotics @LightwheelAI @UnitreeRobotics and more.
5
9
47
3,582
Last day to claim the Day 1 Badge for those who signed up during the BitRobot Bolts App countdown.
Claim expires 11:59PM ET, Sept 21 (3:59 AM UTC on September 22).
4
4
39
3,732
gbot, your @axisrobotics earnings are loaded in the BitRobot Bolts App.
Sign into the Lab, click Profile and connect. Share your IDs below!
Reminder: if you signed up w/ email during our App launch countdown, final day to claim the Day 1 badge is Mon Sept 21, 11:59PM ET
29
13
104
9,693
BitRobot 🦾 retweeted
Looks like our @frodobots Earth Rover Mini got an intelligence upgrade from Muse!
3
11
1,018
BitRobot 🦾 retweeted
We’re building human-centric robotics from the data layer up, the missing piece for generalist robots.
RoboCap + Grounded API turn raw human action into metric 3D labels ready for training.
The hardware and the tooling have always been fragmented. We built them as one stack ↓
10
6
24
3,212
Earth Rover Mini+ spotted in-the-wild with @Meta Muse on board, building its own perception-to-action stack 🤯
Muse was told to find a black ball. So it installed PyTorch + Depth Anything itself, estimated the distance, then drove the rover.
Agentic robotics is getting real.
5
7
31
3,983
Awesome experimenting from @hrhraj, check out the original video here:
Meta's @Muse just drove my robot. I asked, that's it. 🤯
Distance was tricky with only a 2D camera so Muse installed PyTorch + Depth Anything V2 in its own VM, estimated the distance, sanity-checked its own number, then drove there.
An AI agent installing its own tools to act in the physical world. Nice work @AIatMeta 👏
2
9
710
BitRobot 🦾 retweeted
Great way to scale real world data for humanoid robotics
The only way that LLM scaling laws apply is if there is enough data, and this is how you solve the data bottleneck
Congratulations @micoolcho and team!!
1/ Better robot models come from better data.
That takes two things: hardware that captures it fully, and tooling that turns it into training signal.
We built the first. Today, @GroundedSI launches the second: Grounded API. Access both hardware + API ↓
1
4
12
1,005
BitRobot 🦾 retweeted
RoboCaps — foundational data for robotics 🧢
1/ Better robot models come from better data.
That takes two things: hardware that captures it fully, and tooling that turns it into training signal.
We built the first. Today, @GroundedSI launches the second: Grounded API. Access both hardware + API ↓
1
3
13
2,317
BitRobot 🦾 retweeted
Great ego data at scale requires hand tracking to succeed in difficult in-the-wild scenarios, not just in carefully collected settings.
Here are some examples of our hand tracking in the tricky long-tail:
2
16
48
3,819
Grounded API solves a big bottleneck in robot learning: turning raw egocentric video into something usable for policy training.
We're excited to have it part of the RoboCap hardware stack, so buyers get:
> metric 3D hand tracking
> sub-centimeter depth at manipulation range
> SOTA SLAM trajectories
Follow @GroundedSI and comment below for an exclusive launch discount code for the hardware/API bundle.
Grounded API is live.
- SOTA on hand-tracking benchmarks (< 1 cm)
- SOTA on SLAM benchmarks
- In-the-wild ego data -> enriched data in minutes
- Integration with @huggingface @LeRobotHF & @rerundotio
- Built for @BitRobotNetwork RoboCap suite
Technical report & more↓
4
6
29
3,904
BitRobot 🦾 retweeted
Ever since Project Aria came out, I wanted a similar experience that wasn't gatekept by Meta. These guys are doing it. The hardware is legit and API works.
Worth giving them a look!
Grounded API is live.
- SOTA on hand-tracking benchmarks (< 1 cm)
- SOTA on SLAM benchmarks
- In-the-wild ego data -> enriched data in minutes
- Integration with @huggingface @LeRobotHF & @rerundotio
- Built for @BitRobotNetwork RoboCap suite
Technical report & more↓
4
2
71
10,114
BitRobot 🦾 retweeted
Egocentric data is more usable when it’s enriched - and available in formats the community can train on.
GroundedAPI from @GroundedSI turns raw RoboCap video + IMU into millimeter-fidelity 3D motion data, exportable in LeRobot format (Parquet + MP4) and Hub-ready.
Excited to see more large-scale, real-world ego datasets land on the Hub with this release📷
Grounded API is live.
- SOTA on hand-tracking benchmarks (< 1 cm)
- SOTA on SLAM benchmarks
- In-the-wild ego data -> enriched data in minutes
- Integration with @huggingface @LeRobotHF & @rerundotio
- Built for @BitRobotNetwork RoboCap suite
Technical report & more↓
6
19
129
15,660
BitRobot 🦾 retweeted
12k RoboCap units sold to-date.
There'll be millions of hours of ego data collected on RoboCap by our customers/partners and @BitRobotNetwork
1/ Better robot models come from better data.
That takes two things: hardware that captures it fully, and tooling that turns it into training signal.
We built the first. Today, @GroundedSI launches the second: Grounded API. Access both hardware + API ↓
2
5
47
4,239
BitRobot 🦾 retweeted
The amount of egocentric data being collected is exploding. Making all of that footage actually useful for robot learning is where things get really interesting.
1/ Better robot models come from better data.
That takes two things: hardware that captures it fully, and tooling that turns it into training signal.
We built the first. Today, @GroundedSI launches the second: Grounded API. Access both hardware + API ↓
5
2
71
7,005
BitRobot 🦾 retweeted
We are getting to a point where collecting egocentric data won't be stressful anymore, especially the section of turning all that footage into data robots can learn from.
I saw this @BitRobotNetwork RoboCap this morning, which solves a few parts of the data collection process I have been working on. It is built to capture human activity in the real world, and @GroundedSI takes that footage and gives you SLAM, hand tracking, depth and calibration data from it.
You can record someone doing a real task and get back structured data that can go into a robot learning pipeline without having to build every part of that system yourself, which was a stressful process using VR headsets.
Good product.
1/ Better robot models come from better data.
That takes two things: hardware that captures it fully, and tooling that turns it into training signal.
We built the first. Today, @GroundedSI launches the second: Grounded API. Access both hardware + API ↓
4
3
18
1,289