←
Pulkit Agrawal438 posts
Pulkit Agrawal avatar✓

Pulkit Agrawal ✓

@pulkitology

Co-Founder @EkaRobotics, Faculty @MIT

◷ Joined June 2009▧ 129 Photos and videos
Pulkit Agrawal avatar
Pulkit Agrawal@pulkitology·Oct 20

@EkaRobotics has raised a $150M Series A. Generality v/s Performance is the fundamental tension at the frontier of robotics. Generality without performance is just a prototype. Performance without generality is just a single-purpose tool. Customers shouldn’t have to choose. Eka breaks the tradeoff. Powered by our Vision-Force-Action (VFA) model, we deliver the performance and versatility customers need to put robots to work every day. Eka is built on three convictions: ⚡️Force is the language of manipulation. 🚀 Reinforcement learning is how dexterous manipulation scales. 🎂 System 1—fast, reliable physical execution—is the bottleneck to solve. With this funding, we will scale our model and accelerate customer deployments. Grateful to be building alongside a dream team and investors who share our ambition.

0:24
Pulkit Agrawal avatar
Pulkit Agrawal@pulkitology·Nov 5

How do hardware choices affect policy performance?

0:42
Pulkit Agrawal avatar
Pulkit Agrawal@pulkitology·Oct 21

Tactile sensor for any body, now open source that you can build and configure.

1:05
Pulkit Agrawal avatar
Pulkit Agrawal@pulkitology·Oct 19

Manipulation gets solved by feeling; vision isn't as important as you think! A leap in the last mm manipulation that holds back general-purpose robotics 👇

0:13
Pulkit Agrawal avatar
Pulkit Agrawal@pulkitology·Sep 24

Taking humanoid soccer out of the lab: no instrumentation, no controlled environment. Real fields, around people. Trained in simulation. Deployed in reality. No post-training. Just pre-training. Dribbling. Passing. Turning. Scoring goals. Work led by Avi, Khai, @nolan_fey along with @martinpeticco @JohnMarangola, and Venki.

1:37
Pulkit Agrawal avatar
Pulkit Agrawal@pulkitology·Sep 22

Policy learning is not just about learning :)... a lot depends on the physical manifestation underlying what is learned. Uncover things that matter to make robots work!

Post attachment
Pulkit Agrawal avatar
Pulkit Agrawal@pulkitology·Sep 10

Every day there is a new hand—but which one should you use? Despite many attempts to build dexterous hands, there are few efforts to quantify dexterity in the modern era of robotic capabilities. Introducing KaRMA, a metric designed to measure dexterity and guide the design of hand kinematics. Work led by @martinpeticco

0:38
Pulkit Agrawal avatar
Pulkit Agrawal@pulkitology·Aug 22

RNNs are cheap per token, but a fixed state has to compromise across every task it might later face — you compress before you know what matters. With @jyo_pari: dynamic compression. Keep the raw context, and on the first pass, store only enough to know where things are, not everything at full fidelity. When a task arrives, re-scan just the relevant span back through the same recurrence and rewrite the state around what now matters. Takeaway: Storing where to look is cheap; storing everything at full fidelity is not. So don't reread the whole context — train the model to pick the span worth rereading, and let that second look sharpen the state where it now matters.

Post attachment
Pulkit Agrawal avatar
Pulkit Agrawal@pulkitology·May 23

If the goal is test-time search, post-training should optimize for diversity and not just rewards. Introducing Vector Policy Optimization (VPO).

Post attachment
Pulkit Agrawal avatar
Pulkit Agrawal@pulkitology·May 8

Safe human-robot interaction with Eka's force-first approach to robotics: Vision-Force-Action Models (VFA).

0:05