Haptic Labs

@Fabian13Kerj @robotsdigest @QualiaRobotics Thanks for the kind words, Fabian

@robotsdigest Great read. The infrastructure problem is why we’re building @QualiaRobotics.

Robotics lacks infrastructure, not intelligence. Everyone wants to build bigger robot models, but most Physical AI papers complain about the same things: data collection is slow, sim-to-real is fragile, teleop is painful, evaluation is messy, long-horizon control still breaks.

@paul_v_woodward @ai @RoboPapers @chris_j_paxton @micoolcho Interesting. I am very curious how you see: "1, 3, 4 are basically the same problem". I see a large overlap but i separated 1 and 3 was because there was a plenty of data that was NOT manipulation (mobility for one example). I suspect you have a deeper insight I am missing!

@ai @RoboPapers @chris_j_paxton @micoolcho 1, 3, 4 are basically the same problem 2 is the problem being ignored, in favor of successful demo videos. The rest are in service of trying to solve 2 some day.

Haptic scraped all 64 episodes of @RoboPapers (@chris_j_paxton + @micoolcho) and ranked every pain point in physical AI research. The top 10, by mention frequency: 1. Scalable data collection 2. Generalization / zero-shot robustness 3. Dexterous manipulation 4. Teleoperation / whole-body data 5. Sim-to-real transfer 6. Evaluation / benchmarking 7. VLAs / foundation models for control 8. Human video to robot transfer 9. Long-horizon memory 10. RL scaling / offline-to-online Code keeps getting cheaper. Atoms stay expensive. That's the entire startup opportunity in physical AI right now. hapticlabs.ai/blog/2026/03/0…

@jbotsch @ai @RoboPapers @chris_j_paxton @micoolcho Thank you so much for the kind words!

@ai @RoboPapers @chris_j_paxton @micoolcho This is excellent stuff!! Great great write up!

@robotmgz @ai @RoboPapers @chris_j_paxton @micoolcho I like these!

@ai @RoboPapers @chris_j_paxton @micoolcho add to it may be: -tactile sensing - cheap / reliable hardware -autonomous failure recovery

@simonkalouche @ai @RoboPapers @chris_j_paxton @micoolcho My explanations range across: 1. Cynical - "don't need 9s to publish for NeurIPS, but you do to deploy." 2. Structural - "this is industry work" 3. Distribution - "these papers just do not focus on this" 4. Lingo - "they DID talk about it but but buried under assumptions"

@simonkalouche @ai @RoboPapers @chris_j_paxton @micoolcho Totally agree! Fun story: I LOOKED for the "9s of reliability" (or similar) but it was not quite as loud in THESE papers. I loosely suspect it is more a deployment vs research pain (which I will share as well). I say "loosely" because research/industry is blending.

Hot take; it still lacks both

@n6ls0n @robotsdigest Than you for sharing!

Great article: hapticlabs.ai/blog/2026/03/0… h/t @robotsdigest

@robotsdigest Thank you!

very good read: hapticlabs.ai/blog/2026/03/0…

@mynameismattteo @RoboPapers @grok You know that is a good point, matteo... I (@mexitlan) usually listen to articles using substack tools. I can update site so its trivial to listen to it. Good feedback! thank you!

@HapticLabsAI @RoboPapers @grok what is a good tool that I can use to listen to this article on an iphone or mac?

What problems and bottlenecks keep coming up in physical AI research? We analyzed 64 episode transcripts of the brilliant @RoboPapers podcast to see which bottlenecks robotics researchers mention the most. hapticlabs.ai/blog/2026/03/0…

@Theonash_ @RoboPapers Awesome to hear. Would love to see when you have something!

@HapticLabsAI @RoboPapers This resonates pretty heavily with what we’ve been building at mundane. More soon

@RoboPapers 9. Long-horizon and memory (6/64) - Most policies are weak on long sequences and memory-dependent decision making. 10. RL scaling and offline-to-online (6/64) - Exploration, data efficiency, and pushing reliability toward deployment-grade performance remain open.

@RoboPapers 7. VLAs, foundation models, and world models for control (8) - General-purpose models still struggle with reliability, 3D reasoning, and control alignment 8. Human video / human-to-robot transfer (6) - Human demos lack robot-ready actions, dynamics, and embodiment compatibility.

@ai @RoboPapers @chris_j_paxton @micoolcho Thanks for sharing!

Hello world! hapticlabs.ai/blog/2026/03/0…