Learning Legged Locomotion Workshop @ ICRA 2019

Legged robots are notoriously difficult to control. Recent progress in machine learning has shown promises to design robust and agile locomotion controllers automatically. However, most of these learning-based methods are limited to simulation or to simple hardware platforms. Many challenges remain in bringing these learning-based control approaches to real legged robots, including the reality gap, safe exploration, continuous data collection, data-efficient learning algorithms, experimental evaluation, and hardware robustness.

This workshop brings together experts in the fields of legged robotics and machine learning/reinforcement learning to discuss the state-of-the-art and challenges in learning-based control of legged robots.

Topics of Interest

Learning based control for legged locomotion:

  • Reinforcement learning / evolutionary strategies
  • Model based learning
  • Learning in simulation
  • On-robot learning
  • Perception for unstructured terrain locomotion
  • Sim-to-real transfer
  • Hardware platforms for learning
  • Benchmarks
  • State estimation
  • ...

Call for Posters & Robot Demos

Please consider contributing by submitting an extended abstract (1-2 pages). Accepted abstracts will be invited to present a poster during the workshop. We encourage submission of work in progress, experimental hardware results, and "lessons learned" to benefit the community.

Deadline for submission: March 29th 2019. Author notification: April 26th 2019.

Submit your abstract to: learningleggedlocomotion@gmail.com

We also welcome robot demos! Please contact us for more details.

Organizing Committee

contact: learningleggedlocomotion@gmail.com


May 24th 2019 - Room 517d

09:00 - 09:10 Introduction

09:10 - 10:00 Session 1 (2 speakers)

  • Speaker 1 (09:10 - 09:35): Victor Barasuol (IIT): Learning Applied to Modular Locomotion Frameworks
  • Speaker 2 (09:35 - 10:00): Michiel van de Panne (UBC): Learning dynamic locomotion skills for Cassie: an iterative design approach

10:00 - 10:30 Poster session (see below) / Robot demos / Coffee break

10:30 - 12:10 Session 2 (4 speakers)

  • Speaker 3 (10:30 - 10:55): Jemin Hwangbo (ETH Zurich): Sim-to-real transfer of dynamic and agile locomotion policies
  • Speaker 4 (10:55 - 11:20): Erwin Coumans (Robotics at Google): Sim-to-real for quadruped locomotion
  • Speaker 5 (11:20 - 11:45): Shishir Kolathaya & Abhik Singla (IISc, Bengaluru): Realizing Learned Quadruped Locomotion Behaviors through Motion Primitives
  • Speaker 6 (11:45 - 12:10): Deepali Jain (Robotics at Google): Learning complex and agile legged locomotion skills

12:10 - 13:20 Lunch break

13:20 - 15:00 Session 3 (4 speakers)

  • Speaker 7 (13:20 - 13:45): Aaron Ames (Caltech): Learning the Model to Reality Gap in Dynamic Robots
  • Speaker 8 (13:45 - 14:10): Sangbae Kim (MIT): Robots for Robust Physical Interaction
  • Speaker 9 (14:10 - 14:35): Aaron Johnson (CMU): Online and offline learning with contact dynamics
  • Speaker 10 (14:35 - 15:00): Arun Ahuja (Deepmind): Learning hierarchical policies for control of a simulated humanoid

15:00 - 15:30 Poster session (see below) / Robot demos / Coffee break

15:30 - 16:00 Panel discussion (moderator: Jonathan W. Hurst)

Invited Speakers

Aaron Ames (Caltech)

Aaron Johnson (CMU)

Arun Ahuja (Deepmind)

Deepali Jain (Google)

Erwin Coumans (Google)

Jemin Hwangbo (ETH Zurich)

Michiel van de Panne (UBC)

Sangbae Kim (MIT)

Shishir Kolathaya & Abhik Singla (IISc, Bengaluru)

Victor Barasuol (IIT)


  1. Multi-Objective Body Stabilization of a Legged Robot via Distributed Reinforcement Learning: Guillaume Sartoretti, Katayoon Goshvadi, Howie Choset. Poster, Abstract
  2. Geometric mechanics as a seed for learning-based gait design: Baxi Chong, Guillaume Sartoretti, Yunjin Wu, Yasemin Ozkan Aydin, Chaohui Gong, Jennifer M Rieser, Haosen Xing, Daniel I Goldman, Howie Choset
  3. Biologically-Inspired Deep Reinforcement Learning of Modular Control for a Six-Legged Robot: Kai Konen, Timo Korthals, Andrew Melnik, Malte Schilling. Abstract
  4. Learning and adapting quadruped gaits with the "Intelligent Trial & Error" algorithm: Eloïse Dalin, Pierre Desreumaux, Jean-Baptiste Mouret. Poster, Abstract
  5. Lessons Learned from Real-World Experiments with DyRET: the Dynamic Robot for Embodied Testing: Tønnes F. Nygaard, Jørgen Nordmoen, Charles P. Martin, Kyrre Glette. Poster, Abstract
  6. Inverse Optimal Control from Demonstrations with Mixed Qualities: Kyungjae Lee, Yunho Choi, Songhwai Oh
  7. Learning Skills for Humanoid Robots from Video Demonstrations: Jian Zhang, Mario Srouji, Ruslan Salakhutdinov