Aditya Bhatt
@aditya_bhatt
Followers
788
Following
5K
Statuses
434
PhD student (DFKI / IAS / TU Darmstadt) in Robotics 🤖 researching dexterous manipulation with flexible robots 🦾 weird control 🪁 Deep RL 🧠 [email protected]
🇩🇪 ↔︎ 🇮🇳
Joined June 2009
Introducing CrossQ, just published at #ICLR2024! 🎉 CrossQ achieves: 🔥 Very fast off-policy Deep RL 📈 with SOTA sample-efficiency in <5% of the gradient steps 🧹 without target nets or Q ensembles 🧑🏻💻Project and Code: Joint work w/ @DPalenicek 🧵👇
2
16
87
RT @GeorgiaChal: Honored to give an Early Career Keynote at #CoRL2024! My talk, "On the Quest for Robotic Embodied Intelligence," will exp…
0
19
0
0
6
0
Looks like scaling BC is hitting limits in robot learning. We need serious algorithmic improvements to unblock progress.
Twenty six THOUSAND demos for a 75% average success rate on 5 (admittedly hard) tasks. That seems extremely impractical.
2
0
10
@kolibril13 @PyTorch @networkx_team @michelgokan Amazing! You know what would be really cool? Real-time glow visualizations of the edges during a forward pass - maybe modulated by the activations x the outgoing edge weight
0
0
1
RT @aditya_bhatt: One more thing: In a 100% empirical side-quest, we found a surprising alternative to target nets: Bounded activations p…
0
1
0
RT @HaozhiQ: maybe overly intelligent: "Breakthrough in tactile sensing" -- still can't feel Shame
0
16
0
0
14
0
RT @AHendawy19: Are you at @iclr_conf? Come and check our poster on "Multi-Task Reinforcement Learning with Mixture of Orthogonal Experts"…
0
9
0
RT @DPalenicek: If you want to learn about #CrossQ come to our poster session 298 at #ICLR2024 happening right now. Here with @aditya_bhat…
0
5
0
Putting up the #ICLR2024 poster now! Come talk to us at #298 Hall B about how to easily accelerate your RL training with easy network architectural changes!
Now packing 🧳 for #ICLR2024 in Vienna next week! ✈️ If you're into Deep RL for real-world robots, let's talk! 🤖 We'll also be presenting CrossQ's spotlight poster : 🏢 Wednesday 10:45am in Hall B #298 🎬 Talk and slides 📧: aditya@bhatts.org
0
0
5
RT @ias_tudarmstadt: We're presenting five papers at #ICLR2024 this year in Vienna! 🇦🇹 For more details, checkout the thread 👇 https://t.c…
0
3
0
0
4
0
@KagiJournal Did you forget to .detach() the Q(next_s, next_a) predictions before doing backward() on the TD loss?
0
0
3