Mecka
106 posts
Mecka
@MeckaAI
The Data Platform For Robotics Dba Mecka AI
Katılım Ocak 2024
83 Takip Edilen21.5K Takipçiler
Calling all engineers. We're hosting a robotics hackathon during Toronto Tech Week. Humanoids, computer vision, physical AI.
Come build something real.
luma.com/pzyu38qd
English
Mecka retweetledi
Mecka retweetledi
You have no experience.
You’ve never started a company.
You’ve never had a full time job.
Nike is going to kill you.
You’re a kid.
You don’t have technical skills.
You shouldn’t build hardware.
Apple is going to kill you.
You can’t build hardware.
You can’t measure heart
English
Teleoperation was pioneered ~1950 to remotely handle radioactive material. When we use it today to collect robot trajectories for BC, it is still clumsy. Surely, there is a better way! (Hint: human video, RL in sim).youtube.com/watch?v=Iihxza…
YouTube
English
Mecka retweetledi
@MeckaAI (@joshavata , Jason Chong, and team) has been instrumental to the creation of EgoVerse. The team built core infrastructure, including the data browser and capture app, and contributed a major share of the dataset. Excited to keep building together!
English
This project would not be possible without amazing academic collaborators from Georgia Tech, UCSD, ETH, and industry partners from @MeckaAI, @scale_AI @ScaleAILabs, and @RealityLabs @meta_aria.
Huge kudos to students at Stanford working hard on every aspect of the pipeline: Josh Citron, Renee Zbizika, and Jeff Liu. And to project leads @ryan_punamiya, @simar_kareer, @danfei_xu, along with students from Georgia Tech, who did a lot of heavy lifting to bring this project to life 🦾
English
Excited to lead the Stanford effort on EgoVerse and share what we’ve been building 🚀
EgoVerse is a large-scale egocentric human video dataset spanning diverse scenes and tasks, enriched with text and dense hand-pose annotations, and supported by an ecosystem designed for continual growth.
At Stanford, we’ve also been studying human–robot co-training behaviors across several manipulation tasks. A key takeaway: alignment matters.
Strategy consistency is crucial — when human and robot data follow similar strategies, transfer improves; behavioral misalignment degrades performance.
In-domain data is useful — even a small amount of target-domain human data, combined with diverse EgoVerse data, improves visual alignment between the human and robot domain and improves robot policy performance.
Danfei Xu@danfei_xu
Introducing EgoVerse: an ecosystem for robot learning from egocentric human data. Built and tested by 4 research labs + 3 industry partners, EgoVerse enables both science and scaling 1300+ hrs, 240 scenes, 2000+ tasks, and growing Dataset design, findings, and ecosystem 🧵
English
@ryan_punamiya @simar_kareer @anikethcheluva @ryanp_co @LawrenceZhu22 @PranavKuppili @ElmoPA_ @BaoyuLi6 @Randle_Liu @dhruvpatel2012 pleasure working with you all!
English
I also want to take a moment to give a shoutout to our amazing team at GT who gave 200% over the last few months to get this over the line @simar_kareer @anikethcheluva @ryanp_co @LawrenceZhu22 @PranavKuppili @ElmoPA_ @BaoyuLi6 @Randle_Liu @dhruvpatel2012, Aidan Gao, Jenny Chung
Danfei Xu@danfei_xu
Introducing EgoVerse: an ecosystem for robot learning from egocentric human data. Built and tested by 4 research labs + 3 industry partners, EgoVerse enables both science and scaling 1300+ hrs, 240 scenes, 2000+ tasks, and growing Dataset design, findings, and ecosystem 🧵
English
Mecka retweetledi
Thank you to academic partners from GaTech, Stanford, UCSD, ETH, and partners from @scale_AI, and @AIatMeta
Explore at partners.mecka.ai/egoverse
EgoVerse Site: egoverse.ai
Code: github.com/gatech-rl2/ego…
English
Mecka retweetledi
You can use these tools both privately or open-source by contributing datasets in one-click to EgoVerse
The full-stack ecosystem:
- Cloud infra for storage and access
- Mobile and Web Apps for capturing and querying data
- Algos for human-to-robot transfer and deployment
English
Mecka retweetledi
Mecka retweetledi
But accessibility to open human data collection is constrained by specialized wearable devices or missing parts of the stack...
So we introduce the Mecka app on iOS, along with a suite of inference and tools, to reduce the friction from data access to training to evaluation
English
Mecka retweetledi
Robot experiments ran across independent academic partners from Stanford, UCSD, ETH and GaTech all showed a consistent finding...
A small set of aligned human and robot data (same task, same scene) improves knowledge transfer of diverse human data and boosts robot performance
English
Mecka retweetledi
There are two core task types in the initial EgoVerse dataset
- Flagship: that align human tasks and scenes with robot data
- Freeform: captured in-the-wild to show a path to scaling laws
All paired with accurate hand + camera tracking and dense natural language annotations
English
Mecka retweetledi
Mecka is excited to power EgoVerse: a growing ecosystem for robot learning from egocentric human data
Proven across 4 leading research labs, EgoVerse data consistently boosts robot performance
We release the tools for anyone to collect data, inference/train, and contribute! 🧵
English
Congratulations to the entire @MeckaAI team and their research partners
Nobody works harder than these guys
Josh@joshavata
Mecka is excited to power EgoVerse: a growing ecosystem for robot learning from egocentric human data Proven across 4 leading research labs, EgoVerse data consistently boosts robot performance We release the tools for anyone to collect data, inference/train, and contribute! 🧵
English
RT @markgrinev: A year in the making. Proud to have @MeckaAI contribute to this alongside an incredible group of academic and industry part…
English
Mecka retweetledi
This is a massive year-long effort made possible by academic partners from Georgia Tech, Stanford, UC San Diego, ETH Zurich, and industry partners from @MeckaAI, @scale_AI @ScaleAILabs , and @RealityLabs @meta_aria .
Website: egoverse.ai
Code & Data:
English
We couldn't be more excited to have @markgrinev on the team.
Mark has spent years building 0-1 at early stage startups — during his time at Lazer they got to work with some of the incumbents in each domain: Abridge, Polymarket, Decagon, Coinbase.
Now he's bringing that same energy to robotics.
Mark Grinev@markgrinev
excited to announce i've joined my dear friend @jasontheutopian to build @MeckaAI robotics will be the next trillion dollar market and we're building the infrastructure to make it happen more to come soon 👀
English