- Видео 20
- Просмотров 19 553
MPCRL 2021
Добавлен 22 май 2021
The workshop on Recent Advances in MPC and RL for Legged Robots aims to bring together pioneers in employing Model Predictive Control (MPC) and Reinforcement Learning (RL) to control legged robots to discuss the recent advances in the field and the potential ways to combine these two paradigms to have the best of both worlds. We believe that, given the recent advances in both fields, in this workshop we can provide the young researchers with a summary of the state of the art research in the area of legged locomotion control. The workshop is comprised of a series of talks, a poster&demo session and a panel discussion.
Patrick Wensing (University of Notre Dame): Tailoring model complexity in MPC of legged locomotion
The talk was given as part of the ICRA 2021 workshop on recent advances in MPC and RL for legged robots (sites.google.com/view/mpc-and-rl-for-legged-robots).
Просмотров: 1 255
Видео
Nicolas Heess (DeepMind): Towards embodied intelligence
Просмотров 9093 года назад
The talk was given as part of the ICRA 2021 workshop on recent advances in MPC and RL for legged robots (sites.google.com/view/mpc-and-rl-for-legged-robots).
Majid Khadiv (MPI): Model and data, two essential ingredients for controlling legged robots
Просмотров 4783 года назад
The talk was given as part of the ICRA 2021 workshop on recent advances in MPC and RL for legged robots (sites.google.com/view/mpc-and-rl-for-legged-robots).
Daniel Holden (Ubisoft): Robotic characters in video games
Просмотров 1,3 тыс.3 года назад
The talk was given as part of the ICRA 2021 workshop on recent advances in MPC and RL for legged robots (sites.google.com/view/mpc-and-rl-for-legged-robots).
Gerardo Bledt (MIT): Generalizing and improving regularized predictive control for legged robots
Просмотров 8 тыс.3 года назад
The talk was given as part of the ICRA 2021 workshop on recent advances in MPC and RL for legged robots (sites.google.com/view/mpc-and-rl-for-legged-robots).
Jonathan W. Hurst: Learning legged locomotion, RL as one tool in an engineered system
Просмотров 1,2 тыс.3 года назад
The talk was given as part of the ICRA 2021 workshop on recent advances in MPC and RL for legged robots (sites.google.com/view/mpc-and-rl-for-legged-robots).
Jemin Hwangbo (KAIST): Large-scale policy training for robots
Просмотров 1,4 тыс.3 года назад
The talk was given as part of the ICRA 2021 workshop on recent advances in MPC and RL for legged robots (sites.google.com/view/mpc-and-rl-for-legged-robots).
Michiel van de Panne (UBC): MPC and RL, two different roads to legged locomotion, and that's OK
Просмотров 3,6 тыс.3 года назад
The talk was given as part of the ICRA 2021 workshop on Recent advances in MPC and RL for legged robots (sites.google.com/view/mpc-and-rl-for-legged-robots/).
Teaser abstract #15
Просмотров 1133 года назад
In this work, we propose a real-time Nonlinear Model Predictive Control (NMPC) tailored to a legged robot for achieving dynamic locomotion on a wide variety of terrains. We introduce a mobility-based criterion to define an NMPC cost that enhances the locomotion of quadruped robots while maximizing leg mobility and staying far from kinematic limits. A Reference Generator provides trajectories to...
Teaser abstract #9
Просмотров 1253 года назад
In this work, a hierarchical and robust framework for learning bipedal locomotion is presented and successfully implemented on the 3D biped robot Digit built by Agility Robotics. We propose a cascade-structure controller that combines the learning process with intuitive feedback regulations that improve the robustness of the learned walking gait and ensure the success of the sim-to-real transfe...
Teaser abstract 14
Просмотров 1923 года назад
In this work we present a general, two-stage approach for going from a single demonstration trajectory to a robust policy that can be deployed on hardware without any additional training. We show how we can use the demonstration as a starting point, then move on from it to a final policy optimizing directly the relevant task reward and robustness to environment uncertainties. We demonstrate and...
Teaser abstract #13
Просмотров 1253 года назад
In previous works based on MPC for walking pattern generation, only the center of mass (CoM) dynamics are taken into account while the swing foot dynamics are neglected. For generating swing foot trajectories, normally simple polynomial-based trajectories are used. In this work, we show how one can consider a simplified dynamics of the swing foot to write down the swing foot trajectory optimiza...
Teaser abstract #7
Просмотров 923 года назад
In this work, we explore if curriculum learning can improve the performance of trajectory-based policy search on legged locomotion tasks. We evaluate a curricular policy search method that integrates the black-box optimization algorithm CMA-ES (Covariance Matrix Adaptation Evolution Strategy) into a curriculum learning framework that incrementally increases the task difficulty. We assume a dete...
Teaser abstract #11
Просмотров 1383 года назад
This video addresses the problem of computing optimal impedance schedules for legged locomotion tasks involving complex contact interactions.We formulate the problem of impedance regulation as a trade-off between disturbance rejection and measurement uncertainty. We use risk sensitive optimal control that take into account measurement uncertainty and propose a formal way to include such uncerta...
Teaser abstract #10
Просмотров 693 года назад
The simulation of multi-body systems with frictional contacts is a fundamental tool for robotics, computer graphics, and mechanics. Hard frictional contacts are particularly troublesome to simulate because they make the differential equations stiff, calling for computationally demanding implicit integration schemes. We tackle this issue by using exponential integrators, a long-standing class of...
impressive
amazing work!
It is really impressive.
Gerardo, it's been great seeing your RPC work on bipeds! I especially liked seeing the results on Draco 2 in simulation. I'm looking forward to seeing RPC running on the actual Draco2 hardware or the full humanoid. :)
Thanks for the excellent talk Patrick! The reverse mode accumulation approach at 37:30 is a really neat trick. I also really like this new idea of full body MPC at the short horizon first then simple template models at further horizons -- as opposed to one or the other for the entire horizon.
Thanks, Steven!
Thanks, that was interesting!
great talk