Video yükleniyor...
Video Yüklenemedi
I've been working on neural safety systems for robotics. v1 identifies situations as risky/not risky correctly but can't tell between success & failure. Quick wins will come from more data & better labels but the method needs to improve as well
18,453 görüntüleme • 1 yıl önce •via X (Twitter)
8 Yorum

This is a novel method btw, based on mech interp

most notably this method doesn't take backwards reachability into account, which is crucial for HJ & other safety theory. meaning we only consider current state, not future states, for safety my math skills are a bit rusty for this but we'll see if I can work out a theorem 😅

You might need the GVL for success check

This is a cool project but slightly different to what I’m aiming to do here

What’s keeping robot arms from working like human arms? They're big, slow, have the wrong joints, and can't conform to their environment. DexWrist solves all of these issues and simplifies learning constrained, dynamic manipulation👉

“you either work at Tesla/Figure/1X or you don’t work on humanoids” > half a dozen other US humanoid firms > many Chinese firms > many labs have Unitrees > even hackers are buying Unitrees these days the whole thread is like 10% wrong but this stands out as the most wrong.

Humanoid Hiring Strategy Unpacked I analysed 201 robotics job postings from Tesla, Figure, and Agility Robotics. The hiring patterns reveal three completely different approaches to humanoid robots. Here's what you need to know if you're job hunting. ⬇️

Ready for next-gen #robotics?🤖🚗Watch our #AGVs deliver precision and performance.⏰Mark your calendars📢This #mobilerobot will go live on June 30.👏Stay tuned for updates👉 #RobotMobility #AutomationTech #Launch #AI #Newproduct
