Launch HN: Flywheel (YC S25) – Waymo for Excavators
Here's a video: https://www.youtube.com/watch?v=zCNmNm3lQGk.
Interfacing with existing excavators for enabling remote teleop (or autonomy) is hard. Unlike cars which use drive-by-wire technology, most of the millions of excavators are fully hydraulic machines. The joysticks are connected to a pilot hydraulic circuit, which proportionally moves the cylinders in the main hydraulic circuit which ultimately moves the excavator joints. This means excavators mostly do not have an electronic component to control the joints. We solve this by mechanically actuating the joysticks and pedals inside the excavators.
We do this with retrofits which work on any excavator model/make, enabling us to augment existing machines. By enabling remote teleoperation, we are able to increase site safety, productivity and also cost efficiency.
Teleoperation by the operators enables us to prepare training data for autonomy. In robotics, training data comprises observation and action. While images and videos are abundant on the internet, egocentric (PoV) observation and action data is extremely scarce, and it is this scarcity that is holding back scaling robot learning policies.
Flywheel solves this by preparing the training data coming from our remote teleop-enabled excavators which we have already deployed. And we do this with very minimal hardware setup and resources.
During our time in YC, we did 25-30 iterations of sensor stack and placement permutations/combinations, and model hyperparams variations. We called this “evolution of the physical form of our retrofit”. Eventually, we landed on our current evolution and have successfully been able to train some levels of autonomy with only a few hours of training data.
The big takeaway was how much more important data is than optimizing hyperparams of the model. So today, we’re open sourcing 100hrs of excavator dataset that we collected using Flywheel systems on real construction sites. This is in partnership with Frodobots.ai.
Dataset: https://huggingface.co/datasets/FlywheelAI/excavator-dataset
Machine/retrofit details:
Volvo EC380 (38 ton excavator)
4xcamera (25fps)
25 hz expert operator’s action data
We’re just getting started. We have good amounts of variations in daylight, weather, tasks, and would be adding more hours of data and also converting to lerobot format soon. We’re doing this so people like you and me can try out training models on real world data which is very, very hard to get.
So please checkout the dataset here and feel free to download and use however you like. We would love for people to do things with it! I’ll be around in the thread and look forward to comments and feedback from the community!
26 comments
[ 3.5 ms ] story [ 25.3 ms ] threadThe Looming Disaster Under America's Biggest Oil Field [video] - https://news.ycombinator.com/item?id=45361022 - September 2025
Texas has thousands of abandoned oil and gas wells. Who is responsible for cleaning them up? - https://www.texastribune.org/2025/05/08/texas-orphan-wells-e... - May 8th, 2025
The Rising Cost of the Oil Industry’s Slow Death - https://www.propublica.org/article/the-rising-cost-of-the-oi... - February 22nd, 2024
Well plugging SOP:
https://www.epa.gov/natural-gas-star-program/well-plugging
https://www.osha.gov/etools/oil-and-gas/abandoning-well
I've actually spent a decent amount of time running an excavator, as my Dad owns a construction / road building company. It was a great summer job!
An important note about the pilot hydraulics is that they _provide feedback to the operator_. I would encourage any system that moves these controls on behalf of a remote human operator or AI to add strain gauges or some other way to measure this force feedback so that this data isn't lost.
The handful of "drive by wire" pieces of equipment that my Dad or other skilled operators in my family have ran were universally panned, because the operators are isolated from this feedback and have a harder time telling when the machine is struggling or when their inputs are not sufficiently smooth. In the automotive world, skilled drivers have similar complaints about fully electronic steering or braking systems, as opposed to traditional vacuum or hydraulic boosting approaches where your foot still has a direct hydraulic connection to the brake pads.
There was a YouTube video recently of an AI-assisted digger making a wall out of the concrete rubble from a demolition.
I believe the applications for really smart excavators must be huge. Sounds like this might be a step on that voyage.
I ask in all seriousness since, for example, retrofitting regular semis to electric requires millions and millions just to get started
That's because semis - just like cars - are road legal and everything has to be done by the books. The hard part isn't swapping out the engine and mounting a battery pack, car modders have done that for decades for funsies, the hard part is getting it certified for roadworthiness.
On construction sites, particularly ones fully on private property, no one gives a fuck about the equipment, at least not until someone gets hurt, maimed or killed.
I have a Cat 289D skid steer I would happily contribute to the effort if you guys move into the compact equipment space (compact being a relative term, as my machine is only 6 tons compared to your 38 ton machine)
Is that different than what you are doing now?
Hails cab. Excavator shows up
[1] https://youtu.be/QKMsqFMcmL0&t=962
The range of skillset on these things is large, there are 2 dominant (fairly swappable - meaning most machines do both) different operating modes (excavator / backhoe) and I see that operators have a specific one they are best with.
Honestly, when I see a real pro using one of these machines I think this is one area AI is not going to win at soon - in the real world there's a whole support crew working with the excavator operator
This leaves remote operation, which just makes no sense at all. The cost of one guy going to a construction site is never going to be more expensive than retrofitting a fleet of excavators with this hardware and building a remote operating center. Additionally these should obviously not be allowed to be used at construction sites, since remote operation in such a dangerous environment adds a totally new layer of hazards. Direct communication between operators, verbally and visually, is extremely important, to operate an excavator safely.
And last but not least, my car can’t keep lane and can’t drive itself safely in autonomous mode for 50 miles straight on clear day on highway. That’s state of art vision only system in 2025.
The salary of the skilled guy is laughable compared to heavy machinery costs. I was in a quarry on Tuesday. Two guys were operating 7000000€ machinery there, their salary is rounding error in the whole operation. Fuel over the year costs more.
But... I assume you're intending to run on extremely large sites such as highway construction, open pit mines and the likes primarily? Because my experience (if a bit dated cough) is running small 750kg baby excavators under sometimes extreme space constraints - digging trenches for telco in urban and rural areas, which often enough meant having to work with 5-10 cm distance to walls, lighting posts or other infrastructure, and directly next to workers shoving soil into the bucket. Will you add stuff like 360° camera vision, LIDAR etc. to make that safe and help a remote/AI operator, or are you planning on large sites with less danger potential only?
You might be interested in the work of Peter Corke also, he's automated horizontal mine shaft loaders and huge drag line shovels in his research:
https://www.youtube.com/watch?v=YUb9_Ysd2Hw
I think he used a different approach than you do, using visual servoing to get feedback and data from a camera. Maybe there's some value in combining both approaches, learn to control a machine from an operator, and also keep track of what is being moved with a camera to add another layer of control.