Researcher in robotics and machine learning (Reinforcement Learning). Maintainer of Stable-Baselines. Links to find me on other platforms are on my homepage 🦋
Joined June 2018
- Tweets1.3K
- Following338
- Followers3.1K
- Likes3.6K
Pinned Tweet
After several months of beta, we are happy to announce the release of Stable-Baselines3 (SB3) v1.0, a set of reliable implementations of reinforcement learning (RL) algorithms in PyTorch =D!
Blog post: araffin.github.io/post/sb3/
GitHub: github.com/DLR-RM/stable-bas…
Antonin Raffin retweeted
🦋🦋 Hi! I joined the other app... roughly same @ as here (stephmilani). Send me your @ so I can follow you! 🦋🦋
Antonin Raffin retweeted
For those who are interested, the reviews are now publicly available.
#OpenScience
📝 openreview.net/forum?id=ZDvX…
👏 Well done to the whole team. @vwxyzjn @araffin2 @FlorianFelten1 @RousslanDossa @RyanSullyvan and the others
Quick Python tip to improve error messages.
When you throw an exception for an invalid value, give the user some feedback on what exactly that value was.
We tried to follow this principle in our gym env checker: github.com/DLR-RM/stable-bas…
ALT An example of throwing more informative errors. Where name is a tuple, it also prints the variable instead of just complaining it is not a string: f"'name' must be a string, not of type {type(name)}: {name}" vs "'name' must be a string"
A great collection of RL Q&A by James MacGlashan
What is the "horizon" in RL?
Why is the DDPG gradient the product of the Q-function gradient and policy gradient?
And more...
decisionsanddragons.com/
Antonin Raffin retweeted
Reinforcement learning in #AI is hard, so I’ve made a website to collect answers I’ve given to common RL questions.
It's named Decisions & Dragons. It’s launching with 8 questions and answers, but I will add to it in the future.
A 🧵to give a preview with the link below.
Antonin Raffin retweeted
🔊This is an important change all @PyTorch users should be aware of:
`weights_only=True` will now be the default behaviour in torch.load!
torch.load is finally making the BC breaking change and enabling weights_only=True by default dev-discuss.pytorch.org/t/bc…
Antonin Raffin retweeted
Cool demo on upside-down object re-orientation presented by the team from German Aerospace Center (DLR).
#Robotlearning #ReinforcementLearning #AI #CoRL2024
Some news from Stable-Baselines3 (master branch):
- we added support for Gymnasium v1.0 (while being backward compatible with v0.29.1)
- we added CrossQ algorithm in SB3 Contrib (a sample efficient replacement for SAC, without target network)
SB3 v2.4.0 coming soon ;)
Antonin Raffin retweeted
Let’s meet at #CoRL in Munich :)
We are presenting our approach about interactive incremental learning.
If you are interested see the link in the comments or ask me directly:)
I'll be at @corl_conf in Munich this week (with colleagues @GabrielQuere @KnauerMarkus and others)
Don't forget to come and see the in-hand manipulation demo with the Agile Justin robot by @cornerfarmer and others
Antonin Raffin retweeted
🎇 I’m on the academic job market! I’m a PhD candidate at @mldcmu.
My research tackles challenges that arise from the sequential nature of human-AI interaction. Toward this goal, my work involves:
🤖 reinforcement learning,
🧠 foundation models, and
👩💻 human-centered AI.
Shares appreciated!
Antonin Raffin retweeted
Would be extremely useful to open source the model so that people can experience it themselves👌
Robotics needs to go beyond demo videos!
Inspiring work 🤓
Very happy to start sharing our work at Pi 🤖❤️
- a 3B pre-trained generalist model trained on 8+ robot platforms
- a post-training recipe that allows robots to do dexterous, long-horizon tasks
physicalintelligence.company…
What's exciting isn't laundry, but the recipe- a short 🧵
Solving ML Research Problems
A Structural Guide to Approaching Machine Learning Research Problems by @erogol
erogol.substack.com/p/solvin…
Congratulations to the @DLR_en EDAN team and their pilot Mattias Atzenhofer for winning the assistance robot race at @cybathlon 2024
Replay: vimeo.com/1023562261
🏆 And the winner of the Assistance Robot Race at CYBATHLON 2024 is….
EDAN 🎉
Congratulations to the whole team, and a huge thank you to all the ROB teams for taking part! @bfh_hesb @ETH_en
#AssistanceRobot #CYBATHLON2024 @maxonglobal
Antonin Raffin retweeted
One day we are going to wake up, after years of the consistent and pernicious pollution of critical information ecosystems, and deeply regret this era of thoughtless LLM deployment.
A devastating loss of accuracy, clarity and conciseness -- for what?
The internet has been progressively diluted with AI-generated slop. Are medical records headed for the same fate?
🧵 I just published a perspective in @NEJM with @AdamRodmanMD and Arjun Manrai on why rushing AI into medical documentation could be a mistake.
