Carlos Mastalli

Assistant prof. and scientist at @EDINrobotics (@NRobotarium / @HeriotWattUni) and @TheIHMC. PI of Robot Motor Intelligence Lab

[Carlos Mastalli @CarlosMastalli]

How to achieve Athletic intelligence?

In a nutshell, I believe that combing explainable control approaches, efficient numerical optimization aspects, and benefits of data-driven approaches is crucial!
@EdinRobotML @turinginst @MemmoEU

[Carlos Mastalli @CarlosMastalli]

🚀 Our research challenge is now live on the Encode: AI for Science Challengescape.
We’re looking for new AI foundations for safe, transferable motor intelligence—combining physics-grounded optimization with AI for contact-rich robotics.
If this sounds exciting, let’s connect.

[Carlos Mastalli @CarlosMastalli]

@ChongZitaZhang @MichaelAPosa Hype leads to just exploitation.

[Carlos Mastalli @CarlosMastalli]

I am looking for a postdoc that wishes to delve into numerical optimization, contact physics, MPC, RL, and pioneers the next generation of contact-implicit motion generation approaches for loco-manipulation.

For more details see https://t.co/YSwhiexyon.
Thanks!

[Carlos Mastalli @CarlosMastalli]

@_onionesque There are many other interesting directions where we can exploit the structure induced by the isotopic decomposition. Happy to see others appreciate our efforts :)

[Shubhendu Trivedi @_onionesque]

I guess someone had to do it sooner rather than later. Summary: Model dynamics using the Koopman operator. If the underlying system has some symmetry, the Koopman operator must commute with the action, and can be decomposed in terms of the irreps. https://t.co/Gm8IJKFtwn

[Carlos Mastalli @CarlosMastalli]

Again, thank you the organizer for this effort!

[Carlos Mastalli @CarlosMastalli]

Last Sunday, I was delighted to be invited to present our research work on agile and perceptive locomotion in legged robots at the #IROS'23 Workshop on Reactive and Predictive Humanoid Whole-body Control.
https://t.co/SbbaGZR42r

[Carlos Mastalli @CarlosMastalli]

It was a fantastic event organized by Yan Gu, Emel Demircan, Tianwei Zhang and Yuquan Wang discussing the very relevant topic of #humanoid #control.

Below you can see my presentation. However, I invite you to watch other intriguing talks here: https://t.co/t1tgcmrkOE

[Carlos Mastalli @CarlosMastalli]

If you're keen on understanding how we enable inverse-dynamics MPC, then I would suggest watching this 5-minute presentation:

https://t.co/u4ZpfymGMt

If you're going to IROS this year, then feel free to send me an email. Hope to see you next week!
@EDINrobotics @NRobotarium

[Carlos Mastalli @CarlosMastalli]

@Mvandepanne The only thing that I am worried about is the hypes. There is a general unrealistic perception that recent humanoid robotics companies and LLM will enable the type of things in the video. Too much hypes between AI folks and these new companies. Why don’t we listen the experts?

[IEEE Transactions on Robotics (T-RO) @IeeeTro]

Thanks to a cutting-edge model predictive control method developed by researchers @EdinburghUni, the ANYmal robot is defying gravity. Fast, efficient, agile, and even breaking computational speed records - it's like robot Olympics! 🤖🧗 https://t.co/SpzQjk0kQ8 #Robotics #AI

[Carlos Mastalli @CarlosMastalli]

@MichaelAPosa @NSF Congrats Michael! Well deserved

[Carlos Mastalli @CarlosMastalli]

@OlivierStasse I am not concerned about safe and dirty arguments. It is like nuclear energy, you can make it safe with enough attention.

My concern is more about the price of hydraulic robots. It is unlikely to be able to reach the same level of mass production found in electric components.

[Carlos Mastalli @CarlosMastalli]

I am delighted to be part of this lineup. Join us if you're interested in MPC, numerical optimisation, legged robotics and co-design. In my talk, I will share with you our efforts in these topics @EdinRobotML @EDINrobotics @NRobotarium @MemmoEU

[Model-Based Optimization @TCOptRob]

As the new year gets rolling, we are happy to announce that our second seminar will take place on January 13 at 9:30 AM EST, with exciting talks from
Carlos Mastalli and Majid Khadiv. Hope to see you there!

[Carlos Mastalli @CarlosMastalli]

@locoscaron @rohan565singh Indeed, I am also curious about it. Optimal motions using the full dynamics in TO are different. The swing lateral motion is inward, which does make sense to me as this strategy reduces the angular momentum.

[Carlos Mastalli @CarlosMastalli]

These results are cool: https://t.co/Awzn1ROiPu

I wonder what other advantages deep learning bring to boost algorithms for numerical optimisation.

[Carlos Mastalli @CarlosMastalli]

Nice to see more development in terms of analytical derivatives of the rigid body algorithms. Looking forward to seeing this work in Pinocchio:

https://t.co/O86zuswSYx via @YouTube

[Model-Based Optimization @TCOptRob]

The posters for the virtual poster session this Friday are out! Please check out https://t.co/icT4u9lg85

[Carlos Mastalli @CarlosMastalli]

New exciting humanoid robot developed at @IHMCRobotics

See video: https://t.co/V5uqzg7TIT

[Carlos Mastalli @CarlosMastalli]

Last but not least, we have realised this code inside Crocoddyl as we found stimulating to contribute to others: https://t.co/qa29aNGJex

This work has been possible thanks to our excellent team: Saroj Chhatoi, Thomas Corbères, Steve Tonneau and Sethu Vijayakumar.

[Carlos Mastalli @CarlosMastalli]

I am delighted to share with you our preprint titled "Inverse-dynamics MPC via Nullspace Resolution" submitted to an important robotics journal.

pre-print: https://t.co/FdAItWXmh9
full video: https://t.co/gOVFLqvakt

@MemmoEU @EdinRobotML @NRobotarium @HeriotWattUni

[Carlos Mastalli @CarlosMastalli]

Our approach enables the ANYmal robot to climb up damaged staircases, which represents a significant challenge to deploy our robots into the wild.