Resilient Legged Local Navigation: Learning to Traverse with Compromised Perception End-to-End

Under review for: ICRA2024
Paper: arxiv.org/pdf/2310.03581.pdf
WebPage: bit.ly/45NBTuh
Jin Jin* and Chong Zhang*, Jonas Frey, Nikita Rudin, Matías Mattamala, Cesar Cadena, Marco Hutter
(*EQUAL contribution, contact: [e-mail], [e-mail])

[@user-ti2ni8dg7t]

so amazing!!For me, who is being reconstructed by a three-dimensional quadruped robot to navigate the terrain, this work provides me with new enlightening ideas.

[@user-wi5go1op6g]

we can discuss privately and see whether we can get more insights

[@TikiShootah]

Would be kind of funny if the solution to corrupted perception, is add feeler antene to the robot. Come full circle to *bug*

[@MichaelMaravgakis]

Great work and excellent video! I am wondering how the trained policy would react to a sudden change in the slope of the terrain (before the projected gravity changes). For example walking on flat surface and then encountering a slope of 20 degrees (or stairs). Would this be treated as an obstacle if the exteroceptive information is excluded?

[@user-es2zs4br2k]

depends on the locomotion policy and what you simulated. ofc there exists a certain threshold as you can imagine to trigger the "obstacle mode" :P

[@user-wi5go1op6g]

depends on your simulated training environment. The policy actually learning the proprioceptive patterns.