kekeblom

Witty Machines | Senior Robotics Engineer | Zurich, Switzerland | ONSITE | 100-140k CHF + Equity

We build physically intelligent software that enables industrial robots to perform complex manipulation tasks in the food industry - things that were previously impossible to automate. Our systems combine advanced ML, computer vision, and control algorithms to make robots truly adaptable to real-world variations.

We're looking for a Senior Robotics Engineer to help architect and implement our core technology stack. You'll work directly with industrial robots, design perception and control systems, and help shape our technical direction as an early employee.

Tech stack: Python, C++, ROS, PyTorch, industrial robotics platforms

- Good robotics fundamentals (perception, planning, control)

- 1-2+ years hands-on experience with industrial robots or similar systems

- ML/computer vision experience is valuable

- Must be based in or willing to relocate to Zurich

Why us:

- Work on cutting-edge robotics problems with immediate real-world impact

- Significant equity package

- Shape technical architecture and company direction as an early employee

- Central Zurich office location

Apply: [email protected]

At ETH we worked on an analog pressure gauge reading project. In the larger project, we wanted to read analog pressure gauges in oil refineries. https://github.com/ethz-asl/analog_gauge_reader

Aktiia has a product on the market which supposedly already does this https://aktiia.com/uk/. It is based on an optical method though.

The one thing that is painfully obvious to anyone who has rented an electric vehicle in a foreign country, is that charging those things is close to impossible.

Most charging networks require a specific app, that often can only be downloaded in that country's App Store. If you do manage to download the app for a network, signing up can fail because it requires a local address or phone number. Eventually, you'll find a network that does allow foreigners to sign up, but realize that their closest charger is slower and 10 miles away. Often the chargers will show up as available in the app, but actually not be functional and there is no-one at the site to help, even if it's a gas station. They'll just roll their eyes and say it's not their charger. Call the network.

At least gas you can buy with cold hard cash or by simply swiping your credit card at a gas station. No need to input your pronouns into an app.

A good example how society sometimes regresses on some fronts.

These methods learn directly the radiance, in a way which is not lighting dependent. Therefore they would render the same and shadows are not taken into account. Current methods aren’t able to do this, but maybe some day we might discover methods which could infer and take into account some lighting parameters.

In terms of advice, I would for sure start by adding GPS logging into the app. The app logs the ARKit odometry information, which could be a good starting point, but don't know how well that would perform with all that vibration.

It would for sure be a fun experiment, but I think given the extreme conditions, it might be easier to build a custom SLAM sensor rig with a high-quality IMU and several synchronized cameras.

That sounds awesome. I do some mountain biking myself and have sometimes thought that it would be cool if you could replay your rides in 3D through a sparse SLAM point cloud. Of course it would very hard to run SLAM onboard an MTB as there is so much vibration and changes in lighting.

Hey everyone! Here is an app I've been working on over the past couple years for my own purposes. The original idea came from me wanting to collect RGB-D scans for research and development purposes. One option being to buy a depth camera and use that, but I figured I might as well just use the time-of-flight sensor on my phone.

The app has been on the App Store for a while and I originally wrote about it here https://keke.dev/blog/2021/03/10/Stray-Scanner.html. The app can be downloaded here for free https://apps.apple.com/us/app/stray-scanner/id1557051662.

Over the years, tons of users from all over the world have reached out to me asking for the source code, feature requests or adapting it in some way. The users are usually computer vision researchers at universities or corporate research labs or they are engineers working on commercial projects.

Sometimes the requested changes don't make sense for other users. Sometimes they do, but I don't have the time to develop them.

I've just now decided to just release the source code so that people can hack it to do whatever they want with it. Hopefully people will find it useful and that some people might contribute changes back to the app and it becomes more useful for others as well.

Even if it was possible from a legislative perspective, European cities are also considerably harder to drive in than American cities that are cars first with wider lanes and clear grid layouts. Most parts of Europe also have more variable weather conditions.

But yeah, it is indeed also sad that Europe does not have a single credible horse in the race.

I’ve not tried it, but you might want to look into https://www.stack-ai.com/.

I live in Zurich and there are several billboards on my residential street.

There is https://aktiia.com/uk/

I would add that the probability of a task getting done, is inversely proportional to the distance between the people deciding about the task and the people who will actually do it.

Since the board members are almost never the people implementing the decisions, the plan is much less likely to get executed. The decision needs buy-in from the CEO and then the employees. Since the board members don't spend much time with the CEO and employees and they have a very different view of things, that buy-in is likely to be weak, so the plan gets watered down every step of the way.

Really makes you realize why things are so inefficient in public organizations and politics.

In a two person startup, if the founders decide to do something and they are the people executing on that decision, the task is very likely to get done fast and in full. If a non-profit board decides to do something, not so much.

The Financial Times comment section is surprisingly civil and useful despite it’s size. Not sure if that is the kind of community you had in mind.

In case someone is interested in recording RGB-D videos for offline processing, I wrote an app for this https://apps.apple.com/us/app/stray-scanner/id1557051662. It simply writes the raw data files to disk, which you can export to the cloud/your computer.

Working on making computer vision for robotics easier.

I feel like the real reason we don't have robots doing all of our physical grunt work, is that it is just way too expensive to develop robots that do advanced and custom things. Currently only big players with deep pockets can consider developing software for a robot that automates a task that requires some level of fine grained control or visual feedback.

The computer vision part is challenging as you need deep subject knowledge, you need to implement complicated algorithms and assemble huge datasets. We are working on a way to improve the efficiency of creating datasets, standard algorithm implementations and standardizing interfaces to make things as composable as possible.

It ain't much, but it's a start. https://www.strayrobots.io

One thing that could help, but obviously would not completely solve everything is some sort of "organic" label/certification for software. Basically have someone audit the software and check that it doesn't screw the user in any way. For many categories of software, if you could sell the software as not doing any of this or other bad stuff, it would make sense to stay within the bounds of the certification.

Of course some business models are fundamentally toxic and they could never not break some of the rules.

I wrote about the idea here https://keke.dev/blog/2020/11/29/organic-software.html

Yes if you only care about 2d bbox detection, a smart bounding box annotation tool has some advantages. If you need to solve 3D vision tasks, as is the case in 3D bbox detection, 3D keypoint detection, 6D pose estimation, then you need a tool that can also label the z dimension.

Yes it relies on the target being static when capturing the training data, but it’s ok for the background to move. We were actually surprised by how well it works on moving objects without being trained on them. In the post you can see Julius riding on a scooter and that is an unseen example with a detector that was only trained on static scooters.

We plan to charge for some of the algorithms. Also, eventually some parts will run in our cloud and we could charge for compute credits.

Pretty much yeah. Just to be clear, we only use the color and depth images from the camera. There is actually an offline calibration step to obtain camera intrinsic parameters, which are copied into each scene.

The integrate step runs a SLAM pipeline to compute the trajectory of the camera. Then we run an integration step to obtain the mesh.

Our core philosophy is to not stand in the way once you want to do something custom. So totally, if you want to just read the camera poses and 3D labels and do your own thing, you can totally do that and the data is available in each scene folder.

It seems that only Apple can issue refunds on purchases and not developers.

Ah yeah. The app store doesn’t seem to have a way to restrict downloads to only lidar devices. The description does mention the limitation, but there doesn’t seem to be a way to set a hard constraint. So sorry about this! Wonder if there is a way to issue refunds on the app store.

Close! They are actually from Helsinki, Finland.