Header logo is

Shaping in Practice: Training Wheels to Learn Fast Hopping Directly in Hardware




Learning instead of designing robot controllers can greatly reduce engineering effort required, while also emphasizing robustness. Despite considerable progress in simulation, applying learning directly in hardware is still challenging, in part due to the necessity to explore potentially unstable parameters. We explore the of concept shaping the reward landscape with training wheels; temporary modifications of the physical hardware that facilitate learning. We demonstrate the concept with a robot leg mounted on a boom learning to hop fast. This proof of concept embodies typical challenges such as instability and contact, while being simple enough to empirically map out and visualize the reward landscape. Based on our results we propose three criteria for designing effective training wheels for learning in robotics.

Author(s): Steve Heim and Felix Ruppert and Alborz Sarvestani and Alexander Spröwitz
Book Title: ICRA2018
Year: 2018

Department(s): Dynamic Locomotion
Bibtex Type: Unpublished (unpublished)
Paper Type: Conference

State: Accepted
URL: https://arxiv.org/abs/1709.10273

Links: Video Youtube


  title = {Shaping in Practice: Training Wheels to Learn Fast Hopping Directly in Hardware},
  author = {Heim, Steve and Ruppert, Felix and Sarvestani, Alborz and Spr{\"o}witz, Alexander},
  booktitle = {ICRA2018},
  year = {2018},
  url = {https://arxiv.org/abs/1709.10273}