Sammy Sidhu

CEO at @eventualai

Building the frontier of robot understanding with VLMs and data.

Prev: self-driving at DeepScale (acq by Tesla) and Lyft Level 5.

[Sammy Sidhu @Sammy_Sidhu]

IROS was a ton of fun this year, but the papers and posters already feel like last year's research. Deadlines 7 months before the conference will do that.

I know I'm beating a dead horse, but there has to be a better way

[Sammy Sidhu @Sammy_Sidhu]

@isgeles This felt good to see. Too much of hopcopter at IROS

[Sammy Sidhu @Sammy_Sidhu]

Family portrait

[Sammy Sidhu @Sammy_Sidhu]

Siick moves

[Sammy Sidhu @Sammy_Sidhu]

Even the suitcases at IROS are robots

[Sammy Sidhu @Sammy_Sidhu]

I’m at IROS in Pittsburgh this week.

Talk robots with me

[Everett Kleven @everettkleven]

I re-ran VLADBench on 15 of the latest VLMs.

GPT-6 Astra scores 83.6, while Claude-Opus-5.5 scored 82.7.
That's 25.6 points above last year's best from Qwen2.5-VL-72B

Which model is worth running depends on the budget and the task.

[Crusoe @CrusoeAI]

GPU starvation is real: at frontier scale, data pipelines can't feed GPUs fast enough, and idle silicon quietly burns budget.

Crusoe's Eesha Pathak unpacked this with @vantortech's @petewilz, Eventual's @Sammy_Sidhu, Archetype AI's Ivan Poupyrev, and Abundance's Tejes Srivalsan at @autonomousevent in SF at Autonomous 2026.

#PhysicalAI #AIInfrastructure #robotics

[Everett Kleven @everettkleven]

How well do you know the open source physical ai community?

I asked claude to help me craft an awesome list of 178 repos across 11 categories: VLAs, sim, world models, RL infra, middleware, perception, data, evals, and deployment.

https://t.co/HvQ2LPNiYB

[Everett Kleven @everettkleven]

🚢 Daft v0.7.16 has shipped.

🦾 ROBOTICS DATA PIPELINES (and we're just getting started)

> daft.datasets.droid

76k robot manipulation demos, camera feeds, and language annotations. Load them as DataFrames, transform with expressions, feed into PyTorch with .to_torch_dataloader()

22 contributors, 44 changes.

https://t.co/IsiF36CDMt

[Everett Kleven @everettkleven]

🚢 Daft v0.7.15 just shipped.

try_cast() converts types without crashing your pipeline — invalid values become null instead of throwing a runtime error.

Also in this release: LZ4 flight shuffle compression, UUIDv7 partition transforms, PostgreSQL source.

https://t.co/4x8IntMgIb

[Eventual @eventualai]

If a distributed query has to materialize more than a few terabytes of data, there's one operation that will dominate: the shuffle.

Shuffling data at scale has been a real bottleneck for Daft users, so we took the time to fix the root cause and rebuild the shuffle from scratch.

[Jay Chia - eventual.ai @jaychia_]

VLA submissions at ICLR grew 18x in a single year, but World Action Models are showing more promising results when it comes to inference speed and adaptability.

ICLR is the premier gathering of professionals dedicated to the advancement of the branch of artificial intelligence called representation learning.

As Physical AI has gone mainstream, a ton of research has focused on leveraging VLAs to translate the intelligence of LLMs into robotics tasks.

But VLAs are slow, and WAM like Shengshu's MotuBrain achieved 96% on RoboTwin 2.0 with an architecture that supports policy learning, world modeling, video generation, inverse dynamics, and joint video-action prediction in a single model.

"These results show that unified world action models can scale in generality, predictive accuracy, and real-world deployability."

It's crazy that MotuBrain runs at 11 hz and adapts to new humanoid embodiments with only 50--100 trajectories!

Link in the comments

[Sammy Sidhu @Sammy_Sidhu]

@Ken_Goldberg An earlier version of his talk is on youtube:
https://t.co/fROcyPtLOC

[Sammy Sidhu @Sammy_Sidhu]

"GOFE (Good Old Fashioned Engineering) just gets the job done."

The team caught @Ken_Goldberg's talk at @ ICRA 26 in Vienna where he talked about how far robotics has come without deep learning.

Systems engineering, controls, and explicit modeling of geometry & kinematics can take you really far.

#ICRA2026

[Sammy Sidhu @Sammy_Sidhu]

@Ken_Goldberg If you sample thousands of positions and add up your multivariate uncertainties via monte-carlo integration, you can actually come up with a probability that a grasp will succeed.

[Sammy Sidhu @Sammy_Sidhu]

@claudeai design, no need to be so defeatist

[Everett Kleven @everettkleven]

First-class observability in Daft.

Operators, Tasks, Rows, Memory are all surfaced in a dashboard that ships with the install.

+ OTel endpoints for your existing collector.
+ Stuck detection.
+ DAFT_TRACE for console debugging.

~45 PRs across the observability stack.

https://t.co/HWyjBiaePN

[Vikram Sreekanti @vsreekanti]

Lots of exciting news to share today!

1. @RunLLM is now @Herald_Dev. The new name reflects the fact that our AI SRE is the only product on the market that operates autonomously — teaching itself about your product & infra, detecting early warning signs of incidents, and investigating without runbooks. Read more: https://t.co/7GjY5bmsoh
2. Herald was named to the InfraRed 100, an annual list recognizing the most promising private companies defining the future of cloud infrastructure. Thanks to Redpoint for the recognition!
3. We're releasing the beta of the Herald CLI — an agent that runs securely on your laptop and gets up and running in minutes. Sign up for early access here: https://t.co/fbv3byCx2L