cs.RO

LIMBO: Learning and Internalizing Model-Free Barrier Objectives for Agile and Safe Whole-Body Control

Jake Gonzales, Arturo Flores Alvarez, Yu-Ming Chen, Aaron D. Ames, Lillian J. Ratliff, Manikantan Nambi

arXiv: 2609.22075Published: September 18, 2026

AI summary

TL;DR

LIMBO is a framework for synthesizing safety certificates for whole-body control, enabling agile and safe movement. It learns safety values from black-box transitions and provides action-level safety feedback. LIMBO scales to 29-degree-of-freedom humanoid control.

Why It Matters

LIMBO's approach to safety synthesis has practical significance in robotics, enabling more efficient and robust control policies. It can be applied to various tasks, such as dodgeball avoidance and locomotion. This can lead to more agile and safe robotic systems.

Key Findings

learned safety certificates
risk-guided boundary sampling
scalability to whole-body control

Topics

control barrier functions

reinforcement learning

robotics

whole-body control

safety synthesis

Thank you to arXiv for use of its open access interoperability. arxiv.org. aibrief.ai is not affiliated with or endorsed by arXiv.