Darius Burschka

70 min fun in a plenary at IEEE RAS ICRA 2025 LA@Chile Satellite trying to convince people that metric coupling of sensor data is not robust in dynamic environments.

@chris_j_paxton Language model for robotics? Sorry... it just cannot be right.

Backprop through planner is one of my dark horse guesses for what will actually make GPT for robotics work

Current AI just got the internal processing wrong? The systems do not deal with the information but just with labels...Original rule-based AI develops rules processing the actual information, data-driven AI develops clustering techniques for labels. tinyurl.com/ad8wh9b8

@chris_j_paxton Current AI? Never!

How long will it be before end-to-end learning can do this instead of classical robotics? In this video, you can see: - the robot is estimating the location of boxes - it's planning out where its feet will go on future steps - the plans shift as estimation of the boxes' positions changes Looks very much like a classical robotics stack to me, with state estimation and planning and all that. And it's of course one of the cooler videos out there, even now.

Not even this... cats are for sure more intelligent than this #StochasticParrot? It is more a (copy)cat? #AI #llm tomshardware.com/tech-industry/…?

When will the #AI community learn the difference between #knowledge and #intelligence? All these systems (with ChatGPT) keep increasing their knowledge level while having zero intelligence.They do not #understand their outputs. They cannot #reason (use it to solve new problems).

@chris_j_paxton Aren't LLMs an over-kill to be used as language interfaces? The only thing that would justify LLM is intelligence, but they are pretty dumb.

Another capability language models provide *now* - no magic, no handwaving, no hype- is providing a commonsense, intuitive language interface to well tested robot capabilities. Which will be a real game changer

The evolution of #intelligence in #DeepLearning networks?

ZEN moment for the weekend. Do you have the courage to stand in front of a religious group?

@chris_j_paxton Oh, it helps with hype. What sounds better "applied statistics" or AI? 😀😀😀

Renaming Twitter feels really unnecessary to me

I mentioned it months ago that I see a bigger danger in change of models in the same GPT over time, where an initial trust is built to control later. I would no use LLMs for decisions, because only in science you have a chance to spot manipulation. tinyurl.com/envuybks

Wait, it was big news when ChatGPT used 100 million users but it is nothing for any other app? I like this *selective danger reporting". Could it be that more people are using Internet with all these cellphones on public transporta…lnkd.in/ejzKsrN7 lnkd.in/eVgSXxrH

Hmm, this sounds really interesting but not much was in media about it. Existing Diesel engines can be converted to run on a mixture of mostly hydrogen with a rest of 10% of Diesel after modification with high efficiency. I guess i…lnkd.in/eT6E2rXw lnkd.in/eFUdwAwK

I guess that it is time to pay more attention to our Heilbronn Campus from TUM? lnkd.in/en8hS9BY

Oops... lnkd.in/eE6h-BBi

Time for a next trip to Las Vegas. The largest LED screen in the world @Sphere? lnkd.in/ecdVhqRB

Did we play enough with the new toy? "OpenAI may not mind that traffic to ChatGPT is slowing down, as the chatbot is a “loss leader generating sales leads..." It is like the BMW i8, a concept tool that went viral but it was not bu…lnkd.in/eGwnJc-r lnkd.in/eYihXtRM

Known tactics to distract from the your current internal problems? Define a new "external" enemy that you can jointly work against. This prevents people from worrying about current problems. We do not need to wait for some magical…lnkd.in/eMsXv8jz lnkd.in/etHbgsxu

We somehow fear to get behind in a dangerous game with things that we refuse to understand. I would say that at this point it is pretty clear that people driving AI refuse to acknowledge truth about the dangers and necessary improvements to make it safe,…lnkd.in/eNpcKx9Y

We need to rethink our image of AI. Hollywood suggests an almighty, smart system with the best available knowledge. We seem to trade it for a convenient search engine re-producing opinions of an ' Avarage Joe'. Next time you turn to Bart or ChatGPT for…lnkd.in/eKeXY5WM