Our grand finale: A complex, long-horizon dynamic sequence, all driven by a proprioceptive-only policy (no vision/LIDAR)! In this task, the robot carries a chair to a platform, uses it as a step to climb up, then leaps off and performs a parkour-style roll to absorb the landing.

💡 How can humanoids achieve autonomous, generalizable, and lifelike interactions with real-world objects? 🤖 Introducing PhysHSI: Towards a Real-World Generalizable and Natural Humanoid-Scene Interaction System Paper: arxiv.org/abs/2510.11072 Website: why618188.github.io/physhsi/

Warming up for #IROS2025 🔥 We’re releasing CaDeLaC as open source! - Model and training pipeline - Integration with acados for adaptive physics-consistent MPC - Simulation + real Franka torque control 💻github.com/Schulze18/cade… Intelligent Autonomous Systems Group Jan Peters Oleg Arenz 🧵👇

I'm presenting four different works at IROS 2025 this week in Hangzhou 🤖

What if robots could decide when to see and when to feel like humans? We built a system that lets them. Multi-Modal Policy Consensus learns to balance vision 👁️ and touch ✋. 🌐 Project: policyconsensus.github.io 1/N

Robot meets parkour 🤖

By combining learned locomotion with flexible high-level control, Spot is capable of demonstrating dynamic manipulation of objects that exceed its nominal carrying capacity. Check out the full video for more info on how the robot autonomously drag, roll, and stack 15 kg (33 lb)

🚀 Introducing SoftMimic: Compliant humanoids for an interactive future — bringing humanoids into the real world 🤝 🔗gmargo11.github.io/softmimic/ Current humanoids collapse when they touch the world — they can’t handle contact or deviation from their reference motion 😭 😎SoftMimic

Unitree G1 crawl policy deployed to hardware! Plenty of room for improvement, but it's a start.

Very happy to share that "iterated Q-Network (i-QN)" received a J2C Certification from TMLR🥳 What is i-QN about?👇 x.com/Theo_Vincent_/…

Discovering state-of-the-art reinforcement learning algorithms Reinforcement learning agents usually learn with rules we program by hand (TD, Q-learning, PPO…). But humans didn’t hand-design our learning rules—evolution did. What if we let machines discover their own RL update

Very impressive system! Physical stability is a tough yet often overlooked challenge for generation/manipulation. Our prior work (asap.csail.mit.edu) explored planning stable assembly for general geometries. Looking forward to text->3D->dexterous manipulation for anything!

Ever want to enjoy all the privileged information in sim while seamlessly transferring to the real world? How can we correct policy mistakes after deployment? 👉Introducing GSWorld, a real2sim2real photo-realistic simulator with interaction physics with fully open-sourced code.

This is wild: ALL net wealth in the US stock market since 1926 has been generated by just 3.44% of companies. To put this differently, ~97% of all stocks have barely contributed to long-term shareholder wealth creation. The top 1.88% of companies reflect 90% of total gains.

No one: Terri MechKenna: Can't stop, won't stop. We're hiring at @auki btw.

JUST IN: Nvidia is now worth more than: 🇨🇦 2 Canadas 🇮🇸 166 Icelands 🇩🇪 A Germany and a half 🏦 All U.S. regional banks combined 💰 The world’s billionaires, 3× over 🏈 The entire NFL, 35 times 🍕 Every pizza sold in history 🛍️ Every item on Amazon 💳 Visa + Mastercard + PayPal

Following their GAE model launch, Westlake Robotics just showed off a real-time kung fu demo. This means your physical avatar could handle a ton of tasks for you—maybe even step into a boxing match. 😂