Mumuksh Tayal

Working on Physical AI | Reinforcement Learning | Masters Research @ IISc | IIT Gandhinagar '24

[Mumuksh Tayal @MT1699]

@why_always_Neel Hey man, let's connect. I work on Physical AI, currently also building up a project to build my own simulator by fusing some existing physics engine with a forked render engine.

[Mumuksh Tayal @MT1699]

@SlimeTheSystem Would you mind sharing about it here? What's it named?

[Mumuksh Tayal @MT1699]

@sflorimm I can code, but I don't want to go back at this point

[Mumuksh Tayal @MT1699]

@scottastevenson @KyeGomezB Then why are latter ones not capable of that?

[Mumuksh Tayal @MT1699]

@alwayspriyesh Hi, let's connect

[Mumuksh Tayal @MT1699]

@lucyjcai Yes, I guess so. I still don't get it, why language has to do anything with robotics? It takes up most of the trainable parameters anyways

[Mumuksh Tayal @MT1699]

@_AarushiSingh Build a niche to target towards. That would actually help

[Mumuksh Tayal @MT1699]

@piyusch @boknilev Is that good or bad? I mean technically the length of the paper shouldn't be too long that it spills into appendix. For that we have journals, right?

[Mumuksh Tayal @MT1699]

@shiv_ml Quite the progress fr

[Mumuksh Tayal @MT1699]

@vicky_grok One could have studied any open source models, understand them, recreate their own version with custom mods and distill them anyways

[Mumuksh Tayal @MT1699]

@SinghDevHub Degrees are not useless, they add credibility to your profile, but that's pretty much it.

[Mumuksh Tayal @MT1699]

@cq1035344 @charles_maddock That's so accurate

[Mumuksh Tayal @MT1699]

@Google If you have any suggestions or find bugs, please do let me know

[Mumuksh Tayal @MT1699]

Have been working on this project built on top of MuJoCo-Warp, appended with a custom forked @Google Filament Render Engine which I have made to render thousands of environments in parallel on GPU natively.

Checkout the trailer video 👇👇

[Mumuksh Tayal @MT1699]

As it uses Google Filament render engine, you can now drop in any environments in GLB, FBX, etc. formats from your favourite asset stores like @Sketchfab, @polyhaven and add your robots into these ready to use photorealistic environments instantly.

Checkout the project at: https://t.co/dwNvOmeopz

[Mumuksh Tayal @MT1699]

@paraschopra This is very true and a valid problem. And not just that, reviewers at times cite these papers' results during rebuttal, and if you point out the concern that the results don't reproduce, the reviewer questions you for not including this finding in paper.

[Mumuksh Tayal @MT1699]

Also, you can read the full paper here: https://t.co/mPOPYHSL7H

[Mumuksh Tayal @MT1699]

Our paper "Safe Flow Q-Learning" got accepted at @RL_Conference (RLC) 2026! 🚀

RL is growing incredibly fast right now, and I'm just really happy to be contributing to the space along the way.

Looking forward to attending it in Montreal this August! 🇨🇦

[Mumuksh Tayal @MT1699]

@sanskarwtf @iiscbangalore Ekdum