Offline motion libraries and online MPC for advanced mobility skills

Our robot ANYmal combines offline motion libraries and online model predictive control for complex locomotion skills.
Journal article published in the International Journal of Robotics Research (IJRR): journals.sagepub.com/doi/10.1...
Learn more about the robot at www.swiss-mile.com
Video by Marko Bjelonic, www.markobjelonic.com
Title:
Offline motion libraries and online MPC for advanced mobility skills
Authors:
Marko Bjelonic, Ruben Grandia, Moritz Geilinger, Oliver Harley, Vivian S. Medeiros, Vuk
Pajovic, Edo Jelavic, Stelian Coros and Marco Hutter
Abstract:
We describe an optimization-based framework to perform complex locomotion skills for robots with legs and wheels. The generation of complex motions over a long-time horizon often requires offline computation due to current computing constraints and is mostly accomplished through trajectory optimization (TO). In contrast, model predictive control (MPC) focuses on the online computation of trajectories, robust even in the presence of uncertainty, albeit mostly over shorter time horizons and is prone to generating nonoptimal solutions over the horizon of the task's goals. Our article's contributions overcome this trade-off by combining offline motion libraries and online MPC, uniting a complex, long-time horizon plan with reactive, short-time horizon solutions. We start from offline trajectories that can be, for example, generated by TO or sampling-based methods. Also, multiple offline trajectories can be composed out of a motion library into a single maneuver. We then use these offline trajectories as the cost for the online MPC, allowing us to smoothly blend between multiple composed motions even in the presence of discontinuous transitions. The MPC optimizes from the measured state, resulting in feedback control, which robustifies the task's execution by reacting to disturbances and looking ahead at the offline trajectory. With our contribution, motion designers can choose their favorite method to iterate over behavior designs offline without tuning robot experiments, enabling them to author new behaviors rapidly. Our experiments demonstrate complex and dynamic motions on our traditional quadrupedal robot ANYmal and its roller-walking version. Moreover, the article's findings contribute to evaluating five planning algorithms.
Video content:
- 00:00​ Boston Dynamic's dream
- 00:13 Intro
- 00:20​ Dance
- 00:35 Summary
- 01:15 Approach
- 02:47 Outro
Acknowledgments:
This work was supported in part by the Swiss National Science Foundation (SNF) through the National Centres of Competence in Research Robotics (NCCR Robotics) and Digital Fabrication (NCCR dfab). Besides, it has been conducted as part of ANYmal Research, a community to advance legged robotics.
Disclaimer: Robot from ANYbotics; customized by ETH Zürich; strictly for research purposes.

[@my_dear_friend_]

Looks smooth and ready for the roller rink! For promotional rather than practical purposes.

[@arjunps8349]

I’m full of excitement now. This has been “the one” project that inspired me to start working in quadrupeds. I hope that I can work with this team one day.

[@leggedrobotics]

What a compliment. Thanks 🙏

[@Gazzar19]

The movement is so fluent, very impressive!

[@leggedrobotics]

Check out the paper to see how we achieved it :)

[@timoeugster7809]

This is just so awesome! As a student of ETH, this just fuels my passion for robotics and all the opportunities that come with it. Seeing this kind of innovation at my uni, literally 100m from where I‘m studying for the upcoming exams is just incomprehensible…
If my application for the focus project at RSL comes through, I would be able to work on a quite similar project. Fingers crossed haha :)

[@leggedrobotics]

Fingers crossed!

[@DiegoAlanTorres96]

This is seriously astonishing

[@leggedrobotics]

Thanks 🙏

[@TikiShootah]

That is such amazing locomotion. The transitions if you can even call them that
It's like a mammal that's always had wheels

[@leggedrobotics]

This is what happens when toi optimize over the robots full dynamics including the wheels

[@MarcoReis15]

Awesome!!!

[@leggedrobotics]

Thanks!!

[@MonteLogic]

Wow! Floored again!

[@leggedrobotics]

Wr hope that we can repeat!

[@lasertagdreamer]

Amazing! )

[@leggedrobotics]

🙏

[@hughzcl7893]

amazing Jobs！ You set an example to me💪

[@leggedrobotics]

💪

[@louabney]

Raibert is an historic contributor to the field of robotics but in this case the kudos go to the Swiss research organizations

[@leggedrobotics]

What an honor!

[@williamhuang5329]

Hanzhen harmonic drive gear , strain wave reducer,
robot joint , over 30 years experience

[@orpheus696]

you really make me want to join ETH just to have the chance to work at this project! :)

[@leggedrobotics]

Thank you!

[@barbeq9625]

Amazing performance! I've seen the paper but have a question. In 5.2 Torque generation, it is said that u* is translated into desired accelerations and inverse dynamics is used to generate torque, but the torque vector of wheels is not included in the u* since the u_to=[λ qj_dot] is only about contact force and joint velocity of legs. I'm wondering if there is an individual controller using velocity from TO as the control target to generate the torque vector of wheels, or maybe I've missed some important parts?

[@guesmisadok298]

it's realy good project :) amayzing and i'm so interested in this field

[@leggedrobotics]

Thank you

[@ash.ab.5575]

I hope this #SWISS-MILE robot will be the first autonomous MARS explorer. better than old metal wheeled rovers

[@user-hk9eo6ux9h]

where can i buy it?

[@motokokusanagi7683]

Tachikoma!

[@leggedrobotics]

😂