EigenSafe

A spectral framework for learning-based stochastic safety filtering

Inkyu Jang¹, Jonghae Park¹, Chams E. Mballo², Sihyun Cho¹, Claire J. Tomlin² and H. Jin Kim¹

¹ Seoul National University

² University of California, Berkeley

Workshop on Safe and Robust Robot Learning for Operation in the Real World (SAFE-ROL workshop) at CoRL 2025.

cheetah_balance : marked unsafe if the robot flips

w_safety_filter.mp4

EigenSafe safety filter applied

wo_safety_filter.mp4

Reference input (random action)

cheetah_run : marked unsafe if the robot moves leftwards

w_safety_filter.mp4

EigenSafe safety filter applied

wo_safety_filter.mp4

Reference input (random action)

lunar_lander : marked unsafe if the lander touches the ground at an improper (not upright) attitude or moves out of the frame

1.mp4

3.mp4

EigenSafe safety filter applied

2.mp4

2.mp4

1.mp4

Reference input (random action)

4.mp4

Page updated

Google Sites

Report abuse