Akshay L Chandra
@acl21.bsky.social
PhD Student at the Robot Learning Lab, University of Freiburg. Reinforcement Learning, World Models and Robot Manipulation. https://akshaychandra.com
📝 CFP open: #RSS2026 Workshop on Robot World Models 🌐🤖🐨 Submit 4 page novel work or 1 pages abstract of published work! 📅 Deadline: June 5, 2026 More info: 🔗 robot-worldmodels.github.io/submit.html
🚀 Announcing #RSS2026 workshop: Robot World Models 🌐🤖🐨🌐 July 17, 2026 in 🇦🇺! We're bringing together researchers building the next generation of WM for robots — models robots can act in: controllable, long-horizon & grounded in real dynamics. 🔗 robot-worldmodels.github.io
Today I swapped the Franka robot in CALVIN for a human arm and regenerated 6 hrs of "human play" data for free (to quickly iterate over a project idea we have). I don't know what I expected but... it's cute lol 😅
Hyperparameter tuning real-world policy learning by peeking into imagined rollouts inside a world model. Saved us a good few days of iterations. 😅
TIL that dog chew toys make great robot manipulation objects. Colourful, sturdy, durable and full of grip! 🙃
What type of classifier would you use for this data distribution? 🤔🤷♂️
In Germany, I'm obviously an Indian. In India, I'm apparently too German.
Can we extend the power of world models beyond just online model-based learning? Absolutely! We believe the true potential of world models lies in enabling agents to reason at test time. Introducing DINO-WM: World Models on Pre-trained Visual Features for Zero-shot Planning.
Other than CALVIN, are there other tabletop robot manipulation sim environments with a pre-recorded play dataset? 🤔
My IQ temporarily peaked by +200 points just by being here.
Running deep RL on vectorized PyBullet envs that live and render on GPU. It is still not enough to stop our A40 cluster from sweating so much. Maybe I should take the Brax or the PufferLib route... 🤔
i was recently asked to provide 4 "desert island" RL papers. if i were stuck on a desert island i'd hope to have something better to read than #RL papers... but anyway, here's a thread with my choices, maybe you can read them on your flight to @neuripsconf.bsky.social #NeurIPS2024 . Enjoy!
A physical reasoning agent, trained on millions of problems, that you can test out by coming up with physics puzzles and seeing if it can solve them. It's fun to see if you can stump it: kinetix-env.github.io
Investigating large-scale training of RL agents in a vast and diverse space of simulated tasks
kinetix-env.github.io
