Ishan Ahluwalia

Research @ Human Archive | UPenn AI, GRASP lab

[Samay Maini @samay_maini15]

office is on fire but @Ishan_Ahlu is always grinding 🚀🦾

[Ishan Ahluwalia @Ishan_Ahlu]

Everyday at Human Archive we’re doing more to understand and evaluate our data at scale! 🚀

[Raj Patel @babugi28]

We pre-trained the same VLA on 100 hours of Human Archive’s OTS dataset and 100 hours of randomly sampled data from each of @microagi’s and @MeckaAI’s datasets on EgoVerse.

Overall, Human Archive data achieved the strongest results, with the highest overall task success rate. We observed particularly strong generalization to unseen objects.

Thread below 🧵👇

[Ishan Ahluwalia @Ishan_Ahlu]

The PI ecosystem 🤯

[Lachy Groom @lachygroom]

it's finally happening!

[Ishan Ahluwalia @Ishan_Ahlu]

Anyone else have kyren over 50 yds?

[Raj Patel @babugi28]

We’re scaling compute for our perception models and commercial deployments. The team is actively looking for creative ways to partner. Please reach out!

[Ishan Ahluwalia @Ishan_Ahlu]

@VineethSendil @physical_int @agno_three @yuvrajlakhotia I see that makes sense, VLM choice probably did not matter much in that case. Cool project!

[Ishan Ahluwalia @Ishan_Ahlu]

@agno_three Talking like me gang

[Ishan Ahluwalia @Ishan_Ahlu]

@babugi28 this was funny made me chuckle

[Ishan Ahluwalia @Ishan_Ahlu]

@yuvmal1k Send when done pls 🙏

[Ishan Ahluwalia @Ishan_Ahlu]

SO-101 arms are a great, affordable way to get into robotics. Train an end-to-end policy in under 500 bucks. Really recommend.

[Ishan Ahluwalia @Ishan_Ahlu]

@Isaac1385 Yessirrr

[Ishan Ahluwalia @Ishan_Ahlu]

@chris_j_paxton Agreed : )

[Ishan Ahluwalia @Ishan_Ahlu]

Really excited about this release. Doing proper evals of large scale data is often overlooked and it was amazing seeing the effort we put into high quality data having direct benefits on robotic policies. Looking forward to sharing more research soon 🚀

[Raj Patel @babugi28]

We benchmarked only 7 hours of Human Archive data against 4 widely used public datasets totaling 17 hours: HOT3D, H2O, HOI4D, and TACO.

The results:

Human Archive only: 85% task success | 65.6% task progress
4 public datasets: 57.5% task success | 52.5% task progress
Mixture of all 5: 60% task success | 56.3% task progress

[Ishan Ahluwalia @Ishan_Ahlu]

It feels like in the last month, every week has been a massive week for robotics. It's amazing to see the rapid progress and early signs of robotics having a real impact in the world : )

[Ishan Ahluwalia @Ishan_Ahlu]

This is the fastest I've seen robotics in the real world.

[Reward AI @RewardAI_]

Introducing OM-1, our first robot foundation model, zero-shot generalizing to any robot: table-top arms, industrial arms and humanoids.
- learned directly from human manipulation data
- no teleop/robot data
- close to human-level dexterity and efficiency
- multi-robot collab

[Ishan Ahluwalia @Ishan_Ahlu]

@limbs_long Or a guy who is going through education with AI has an opinion that there needs to be a change in education and cares about the future, so he voices it?

[Ishan Ahluwalia @Ishan_Ahlu]

@cobracommanduhr Waymo (10x safer, 94% fewer serious car crashes)

[Ishan Ahluwalia @Ishan_Ahlu]

@cobracommanduhr alpha fold

[Ishan Ahluwalia @Ishan_Ahlu]

@cobracommanduhr Moderna’s new personalized melanoma treatment. AI is used to analyze each patient’s tumor mutations and select the neoantigens used to design their individualized cancer therapy. It just succeeded in Phase 3. Happened like a month ago