Robot trainer (Student Researcher) @StanfordAILab working with @chelseabfinn
Also love (Astro)physics, History, Economics, Math. Stanford, CAJoined June 2020
How should an RL agent leverage expert data to improve sample efficiency?
Imitation losses can overly constrain an RL policy.
In RL via Implicit Imitation Guidance, we show how to use expert data to guide more efficient *exploration*, avoiding pitfalls of imitation-augmented RL
Not all human-collected demos are created equal:
✔️ All are successful
❌ But some strategies are unreliable or brittle
This can hurt final performance.
Demo-SCORE self-curates reliable training data using online experience.
Paper and videos: anniesch.github.io/demo-score/
Data curation is crucial for post-training recipes.
But how do we curate?
Curation is usually manual & tedious. And, it's hard to tell if a strategy in the data will be reliable!
We introduce an automatic way to curate, informed by the robot's policy learning.
Data curation is crucial for post-training recipes.
But how do we curate?
Curation is usually manual & tedious. And, it's hard to tell if a strategy in the data will be reliable!
We introduce an automatic way to curate, informed by the robot's policy learning. https://t.co/10ISDvgwgr
Very excited to introduce ROAM, our new work that allows a robot to *adapt on-the-go* as it faces OOD situations during deployment, drawing on pre-trained behaviors.
See as ROAM enables our Go1 to roller skate zero-shot 🤖🐕🛼 (without any lessons!)
🧵(1/9)
Reviewing #ZooGems@NASAHubble data, I was struck once again by the sense of depth in this view. The lower band of stars is the edge-on galaxy UGC 12281 at 110 million light-years, barred spiral is at 1.5 Gly, edge-on spiral still beyond. Galaxies beyond galaxies beyond galaxies.
Henri Poincaré, summarizing how I feel about mathematics: "It is by logic that we prove, but by intuition that we discover."
("C'est par la logique qu'on démontre, c'est par l'intuition qu'on invente.")
My heart has always been with the intuition.
We're all spun up! Get up (or stay up) and join the #MarsHelicopter team live in mission control to find out if the first powered flight attempt on another planet was successful.
Mission Control Live
April 19 at 3:15am PDT (6:15am EDT/1015 GMT)
youtu.be/p1KolyCqICI
2K Followers 1K FollowingBuilding 🤖... Prev: Founder/CEO at Flock (acq. by Paychex). Disciple of the outdoors, lover of fine foods, dogs and westerns. Have bag - will travel.
123 Followers 845 FollowingInterested in AI, climate and politics. Just beginning a DPhil on world models @oxfordrobots. Prev MSc @ucl, @tortoise, Physics @UniOfOxford
85 Followers 290 FollowingFounder/CTO at AKASA. Previously Google research scientist, Founder of Numovis (sold to Google) and founder/CEO of Cloudlabs(sold). Created ProHDR.
9K Followers 459 FollowingThe Department of Electrical Engineering & Computer Sciences at UC Berkeley. Pioneering the frontiers of science & technology with broad impact on society.
123 Followers 845 FollowingInterested in AI, climate and politics. Just beginning a DPhil on world models @oxfordrobots. Prev MSc @ucl, @tortoise, Physics @UniOfOxford
2K Followers 1K FollowingBuilding 🤖... Prev: Founder/CEO at Flock (acq. by Paychex). Disciple of the outdoors, lover of fine foods, dogs and westerns. Have bag - will travel.
85 Followers 290 FollowingFounder/CTO at AKASA. Previously Google research scientist, Founder of Numovis (sold to Google) and founder/CEO of Cloudlabs(sold). Created ProHDR.
882K Followers 52 Followingwe invest in software eating the world
https://t.co/A9eTFq6plZ
https://t.co/MXGUBJoesw
Watch "The Ben & Marc Show": https://t.co/eRuDhx7kpe
4K Followers 1K FollowingAccelerating aligned AI & a flourishing future with neglected approaches & AI R&D. CEO at @aestudiola (AI consulting co puts profits into AI frontier)
567 Followers 50 FollowingPhD in Computer Science candidate @Stanford. My interests span learning for robotic planning, control, vision systems, and their interfacing representations.
3K Followers 1 FollowingRecording daily paper notes by @chongzitazhang who is interested in robot learning and control, with special focus on legged robots.
///I'm picky///
No recent Favorites. New Favorites will appear here.