Gabe Margolis
@gabe_mrgl
Followers
930
Following
932
Media
7
Statuses
86
Robotics at @MIT_CSAIL
Cambridge, MA
Joined April 2021
Check out our new work on bridging sim and real for athletic tasks!.
Excited to share my recent work with @gabe_mrgl , Martin Pettico, and @pulkitology . We’re pushing the limits of whole-body control to make robots faster, stronger, and more athletic!
0
0
38
I think it connects to this thread. How many "bits of information" in the policy are/should be hard-coded with structured priors? . It's well-motivated to include some if sim != real. But ideally they're "general" ones that apply across many tasks.
If there's not that many bits to learn, then researcher input becomes non-negligible. "I found a trick that makes score go up!" -- yeah, you just hard-coded 100+ bits of information; a winning solution is probably only like 1000 bits. You see progress, but it's not the AI's.
0
0
2
Interesting points about the evolution of sim2real.
1
0
5
Cool work from @TifannyPortela ! . RL is great for robustness. But when it comes to learning whole-body 6D reaching, it takes careful task design to also be *really precise*.
Excited to present the 𝗳𝗶𝗿𝘀𝘁 𝗽𝗮𝗽𝗲𝗿 𝗼𝗳 𝗺𝘆 𝗣𝗵𝗗 ! 🎉. Let's chat if you're attending #ICRA2025, I'll be presenting:.📍Mon, May 19 | 9:40-10:40 — Workshop on Field Robotics.📍Thu, May 22 | 08:30–09:00 — Session ThAT5. @ETH @ETH_AI_Center.
0
0
7
Catch @nolan_fey's oral/poster at today's ICLR Robot Learning Workshop!. The paper will also appear at RSS.
Excited to share my recent work with @gabe_mrgl , Martin Pettico, and @pulkitology . We’re pushing the limits of whole-body control to make robots faster, stronger, and more athletic!
1
5
35
RT @pulkitology: Presenting Unsupervised Actuator Nets (UANs) that push the limits of agile whole-body control without the need for reward….
0
19
0
RT @ericliuof97: Introducing HugWBC: A Unified and General Humanoid Whole-Body Controller for Fine-Grained Locomotion!.Project: https://t.c….
0
14
0
RT @kevin_zakka: The ultimate test of any physics simulator is its ability to deliver real-world results. With MuJoCo Playground, we’ve co….
0
180
0
RT @ZhangWeiHong9: [1/4] 🚨 Excited to introduce Embodied Red Teaming (ERT) – an approach based on vision-language models (VLM) to automatic….
0
4
0
RT @haoshu_fang: A good hand can push intelligence development. Introducing Eyesight Hand, equipped with full-hand high-res tactile sensors….
0
45
0
RT @marceltornev: Robot learning is fundamentally limited by data – human teleoperation on real robots is expensive! 🤖👨💻We propose an alte….
0
27
0
MuJoCo Simulation + Vision Pro AR = 🎯DART 🤖😍🥽.
Collect robot demos from anywhere through AR!. Excited to introduce 🎯DART, Dexterous AR Teleoperation interface enabling anyone to teleoperate robots in cloud-hosted simulation. With DART, anyone can collect robot demos anywhere, anytime, for multiple robots and tasks in one
0
0
6
RT @BostonDynamics: Atlas is autonomously moving engine covers between supplier containers and a mobile sequencing dolly, using ML to detec….
0
389
0
RT @davide_tateo: I'm happy to announce that we will organize the LocoLearn Workshop at @corl_conf 2024 in Munich!. If you are interested i….
0
12
0
Our ICML position paper! We argue that to generalize existing successes of reinforcement learning for robotics, we need to address the interaction between environment shaping & the dynamics of learning. w/ @younghyo_park @pulkitology.
Presenting our #ICML2024 Position paper: “Automatic Environment Shaping is the Next Frontier in RL”. We argue that more reliable and principled techniques for RL environment shaping will pave the path towards generalist robots 🧵 [1/n] @gabe_mrgl . Oral talk happening today
0
2
20
RT @taochenshh: Got cool dexterous manipulation results? 🤖🔥 Consider submitting a short paper or just a demo video to our RSS Workshop!. 📚L….
0
7
0
RT @pulkitology: Make a difference in robotics by helping us identify and select papers pushing the frontiers of robot learning for the upc….
0
21
0
RT @abhishekunique7: Excited about @ZoeyC17's new work on real2sim for robotics! We present URDFormer, a technique to learn models that go….
0
23
0
RT @haoshu_fang: Generalization is always an important topic for robotic manipulation. RISE is a 3D imitation learning framework that demon….
0
6
0
Our #ICRA2024 work on sim2real loco-manipulation! . Instead of training to track gripper position commands w/ undetermined stiffness, we train to track desired forces. This lets us teleoperate with behaviors like impedance control / gravity compensation. Led by @TifannyPortela.
Check out robots using their entire body to increase their workspace and the force they apply. We enable explicit whole-body force control to expand the ability of mobile quadrupeds. Find details: Work led by @gabe_mrgl @TifannyPortela . At
0
6
40