Learning Quadrupedal Locomotion over Challenging Terrain

We present a radically robust locomotion controller for quadrupedal robots which is trained using reinforcement learning in simulation. It uses no external sensing like cameras or Lidar and relies only on internal sensors like IMU and joint encoders.
Learn more about this work on our project website:
leggedrobotics.github.io/rl-b...
Paper links:
- Science Robotics: robotics.sciencemag.org/conte...
- Author's version: arxiv.org/abs/2010.11251
Video by Joonho Lee

[@thesuperiorman537]

Wow, impressive it was achieved with only lMU and encoders. Guess lidar/camera were used for the higher level robot trajectory planning and terrain adaptation depends on your solution! Awesome work! Looking forward to your paper.

[@tonylouis2742]

RSL is my dream lab

[@JackSPk]

TCN = Temporal Convolutional Network

[@nithinvasishta6407]

Climbing down stairs using only proprioception.. that's quite amazing

[@shagui]

Very good explanation and video!

[@ArifRahmanMY]

Wow! Now we’re talking! Or rather walking 👍👍👍

[@ToadalChaos]

Truly next level stuff.

[@yinghaogao3410]

Impressive work !

[@richardlovenvidiagb4897]

Excellent work.

[@stevenhong7099]

Excellent work!

[@anibalasubramaniam4153]

Impressive! I'd love to see the source code of your project if you ever end up publishing it. Keep up the awesome work!

[@weisserelephant]

This is crazy: Love it. I'd like to do this as well!

[@heinrichwonders8861]

Impressive! What is the next step?