@ahadj0

building autonomous robots | @spacex | prev @aws @amazon

ahadjawaid.com
Joined October 2022
  • Tweets1.2K
  • Following171
  • Followers1.9K
82Media
Pinned Tweet
ahad
@ahadj0Jun 29
what if we treated robot actions like audio waveforms? introducing neural action codec (NAC) a method for creating a compressed action vocabulary to train VLA policies
20
28
14
409
57,773
the current step of robotics is figuring out a scalable robotic training recipe that can scale to widest distribution of tasks and embodiments
3
244
yup, suprisingly audio codec models are good action tokenizer recipes! will presenting this work at corl in november!
Trossen Robotics@trossenroboticsSep 25
Two of our arms striking a match, fully autonomous. Ahad Jawaid pulled this off for his CoRL 2026 paper on our previous generation arms. Our 2026 lineup is built for even more dexterous work, with academic discounts for research labs: buff.ly/nNrnn7K
4
8
133
18,208
i’m going to be presenting this work at CoRL in Austin on novemeber 9-12th! would be cool to chat with people who are working on scaling robot foundation models and multimodal tokenization
ahad
@ahadj0Jun 29
what if we treated robot actions like audio waveforms? introducing neural action codec (NAC) a method for creating a compressed action vocabulary to train VLA policies
2
7
55
7,839
Starmind spacex.com/spacexai/starmind…

SpaceX

SpaceX designs, manufactures and launches advanced rockets and spacecraft.

spacex.com
4,370
2,272
266
13,285
3,283,000
i joined @SpaceX ! we need self replicating autonomous robots to make us a multi-planetary civilization
ahad
@ahadj026 Jun 2025
we will need robots building robots if we are going to have a space civilization. this is true even with agi or asi
79
38
5
1,433
118,896
introducing hierarchical implicit q-chunking (HIQC) long-horizon offline reinforcement learning is hard the more steps an agent takes, the more value estimation errors compound i reduce this by decomposing the horizon into smaller segments with subgoals and action chunks
3
11
96
7,102
more replies
this simple idea leads to an increase of nearly 10%+ average compared on 9 long horizon simulation task xitter.cf/ahadj0/status/20723344…
2
1
8
1,099
1
4
918
i found by simply using a high level policy predicting latent subgoal and a low level policy prediciting action chunks you can reduce the number of bootstrapped value estimates to O((T/k)^(1/2)) xitter.cf/ahadj0/status/20723344…
2
1
8
1,088
2
586
i’d like to thank @seohong_park for open sourcing his work as this work was heavily based on those prior works project: ahadjawaid.com/hiqc paper: ahadjawaid.com/hiqc/59_Offli…

Offline RL with Hierarchical Action Chunking (HiQC)

HiQC (Hierarchical Implicit Q-Chunking) learns goal-conditioned policies from offline data: a high-level latent planner plus k-step action chunks and chunk-conditioned critics for deeper unbiased...

ahadjawaid.com
6
508
Check out Ahad's work on leveraging neural audio codec architectures for VLA action tokenization. An interesting direction that adapts ideas from audio compression to represent continuous robot actions.👇
ahad
@ahadj0Jun 29
what if we treated robot actions like audio waveforms? introducing neural action codec (NAC) a method for creating a compressed action vocabulary to train VLA policies
2
54
9,832
here is a policy comparison to a few other methods
1
1
20
2,044
i will like to thank the authors of the OAT paper as our codebase was developed on top of theirs @liu730chaoqi and my co author @YuXiang_IRVL for making this work possible paper: arxiv.org/abs/2606.21372 code: github.com/ahadjawaid/nac project page: ahadjawaid.com/nac

NAC: Neural Action Codec for Vision-Language-Action Models

Vision-language-action (VLA) models rely on discrete action tokenizers to bridge continuous robot control and autoregressive sequence modeling, yet existing tokenizers often trade off between...

arxiv.org
3
36
1,918
introducing lattice a low-cost robot arm with a modular, detachable gripper so the robot can swap its own tools the adapter is $14 if you already have an so101 arm, and about $140 to build the full system from scratch it’s now fully open source
39
42
10
521
40,720
more replies
also want to acknowledge @jeluna20 as a core contributor on this project, and the open source repos it’s built on: @lerobothf @uma_robots @vectorwang2
1
11
1,012
repo: github.com/ahadjawaid/lattic…

GitHub - ahadjawaid/lattice

Contribute to ahadjawaid/lattice development by creating an account on GitHub.

github.com
3
1
1
22
1,043
Load more