TUM dropout. Building machines that build anything.
Joined July 2025
- Tweets 632
- Following 689
- Followers 1,264
- Likes 2,705
112 Photos and videos
Pinned Tweet
May 4
Anyone who is looking for Assembly as a service done by Robotic cells?
Let me know. DMs are open.
(currently based in germany. Nextis offers Aaas worldwide)
1
4
47
5,126
Jun 12
was quiet for a couple of days. But I’m back!
Developed and improved the pipeline to set up a new task. 😏
Disassembly is cool for continuous operation together with the normal assembly.
This task was much faster to set up and also much improved policy performance from the start!
Maybe i can do a continuous loop showing assembly and disassembly by next weekend as a stream
4
10
80
4,963
Jun 1
new top lights arrived. as soon as the side and top frosted glass plates arrive they will be installed for bright diffuse light! (should be tomorrow)
3
33
1,509
May 24
from now on I can replay my recorded datasets in simulation or simulate the policy outputs.
Major key for safety and validation!
will post a video soon 👀
5
4
49
1,842
May 21
everyone benchmarks success rate. the metric that actually determines economic usefulness is TTLNT
-> time to learn new task. and the real unlock is when the policy is substrate-independent: doesn't care which cell it runs on.
3
1
10
805
May 21
With my custom architecture I now achieve on average 60% progress on my gearbox assembly task.
The original average progress of pi0.5 base (after 10k fine tuning) was 14% .
Trajectory of progress is looking good. Now have to reduce the cycle time of iterating on a base policy.
10
7
141
10,941
May 14
Spun up an auto-labeler for my robot intervention datasets.
Labels task stages, aligns negative/correction spans around human takeovers, and drops everything into a
reviewable video timeline before touching the dataset.
Manual labeling was becoming the bottleneck. This should 10x the iteration loop.
8
572
May 14
every time in am in awe when these two go to work.
Currently codex is employee of the month at Nextis but cannot wait for them to get the title.
33
1,606
May 10
started another shift training for the night.
Will hopefully post a cool video of the policy tomorrow….
9
422
May 10
I will build a UMI gripper for quick additional data collection for my cell.
Stay tuned in the next couple of days.
1
47
1,819
May 10
collection another intervention dataset rn. The first run actually showed to improve performance of the base model.
Let’s see if I can keep the flywheel going and really push performance.
Doing delta command intervention takes some getting used to, but it’s pretty smooth overall.
8
489
May 4
once again adding metadata. The first try of SHIFT did not really succeed in improving the policy.
Will collect more intervention data and do another training run, with much improved advantage labelling and segmentation.
Gone is the arbitrary deque rollback buffer.
The night is young
2
19
1,256
Apr 29
Quality time with shawty: we're both crunching some good ole numbers.
4
20
927